Even though in these cases it doesn't matter because of version exclusivity, keep them ordered by type instead of grouped by version.

Decoded some unknown fields and blocks for WorldShift and Guild 2.

NiSourceTexture is now correctly decoded.  The unknown byte actually controls whether the image data Ref gets serialized or not, but it was always 1.  The addition of another conditional value increases the complexity for Pixel Data to 3 rows, all exclusive by version or condition.

NetImmerse NiParticle (called NiOldParticle after Gamebryo 1.1) was severely undecoded.  This affects several NetImmerse games like Morrowind and Freedom Force.

The per-particle info in NiOldParticle actually remained identical to the class in NiParticle, which is named "NiParticleInfo".  It did however have a different name in NetImmerse which was "NiPerParticleData".  Since they were the same data, the two compounds were merged and NiPSysData was updated.

The way that BSPackedCombinedSharedGeomDataExtra refers to the shared geometry was decoded.  It stores the hash of the filename and the offset to the data.

For #70 and #76

For #70 and #76

Added tokens for range attribute strings.

Did initial pass of ranges, many from Bethesda's official export scripts for FO4, which limit the values in the UI. Also have begun using data analysis to find implied min/max ranges (as well as sane defaults).

BSGeometrySegmentData - Combined rows which did not be separated by version.  Adopted better naming from FO4.

AspectFlags type for NiParticlesData for BS.

Changed the unused W component of what should be a Vector4 for Vertex to 'Unused W' for all rows.  Should seriously consider making these all Vector4s for endian correctness.

Misc comment cleanup.

Specify SizedString vs NiFixedString where appropriate.  The compound type `string` is an extra level of indirection in many cases, where the block can never be used on versions less than 20.1.0.3 where the CString to NiFixedString transition happened.

NiFixedString is a basic type, so using it where possible is recommended over the `string` compound type.

`string` could NOT be changed when dealing with multiple rows with the same name but version exclusive, such as "Accum Root Name".

Changed ShortString to ExportString as it was already documented as being exclusive to the Bethesda stream header.

Added SizedString with ushort length type, for the SSF File path.

For #76

Implement initial subclass defaults.  Had to add additional attribute to deal with versioning of defaults, like block versioning.

Added more defaults to normal fields.

Added default to Vertex Color array, also enforcing that members with `arr1` can have a `default`.

Improved class descriptions and fixed Flags values and defaults.

For #72

So that first party modules can be singled out if needed.

For #72

Renamed modules from Gamebryo:  NiParticle, NiPSParticle.

As of Gamebryo 2.5, old NiParticle was renamed NiParticleOld and the new NiMesh-based particle system was named NiParticle.  I didn't like this naming scheme given NiParticleOld is more important to us.

Ni Modules not in Gamebryo:  NiCustom, NiLegacy

These are to handle third party blocks and blocks from NetImmerse (pre-Gamebryo).

Ignored from Gamebryo:  NiCollision, NiPortal, NiPhysXParticle

These were instead absorbed into other modules.

Bethesda modules:  BSMain, BSAnimation, BSParticle, BSHavok, BSLegacy

Animation, Particle, Havok, and Legacy are not modules used by Bethesda.  Legacy is for Morrowind-specific blocks (NetImmerse).  In addition,  BSShader is a module for all the shader related blocks and that has all been folded into BSMain.

Second default values pass, plus some general changes to accommodate them.

The correct name for StringIndex is NiFixedString.  Also added NiTFixedStringMap which is used once in the NiPhysX module.

Also use `true` instead of `1` to match other XML booleans.

Made a correct bitfield using `uint64` as the underlying type, split into nibbles for the sizes and offsets of the vertex attributes.

Changed VertexFlags to VertexAttributes and it no longer has the empty nibble in front.

ARG passing of the BSVertexDesc is treated as the underlying numeric value of the bitfield, so that the attributes can be passed into BSVertexData using `#RSH# 44` instead of the member accessor syntax.

This makes the first use of the basic `uint64` type, so (de)serialization now requires support of 64-bit values.

Cleaned up class naming e.g. Descriptor->CInfo to match the engine and other classes already named that way.  Added hk/bhk to front of type names where appropriate.

Changed Info/Property names to be consistent e.g.  "Rigid Body Info", "Entity Info" instead of CInfo.

Changed all "Unused" fields to a consistent naming scheme, using byte array types and binary="true".

Class hierarchy fixes.  Class documentation complete overhaul.

Some existing naming collides with new naming of different fields, e.g. bhkMoppBvTreeShape "Scale" used to be "Shape Scale" and now "Scale" is  "MOPP Code" -> "Scale".

The previous decoding was all guesswork, though did align somewhat.  Enough to know that the decoding from FO4 applies to Skyrim as well.

Added `default` to a large number of members, as well as `calc` to a few members to enforce array sizes based on other data.

Also fixed a few `type` issues.

This is a first pass in a much larger effort,  using data analysis to gather the most commonly used values.

FO4 changed the Color3 to a Color4 for Skin Tint.  Leaving as separate Alpha for now however.

Also found out the point of the BSSkin::Instance array which is non-uniform scaling.

For Name-based APIs, naming all NiBound fields consistently helps with generic treatment of NiBound getting and setting for any NiObject.

Split up the Rigid Body CInfo into multiple compounds for each Havok version (though left 550/660 combined).  Followed suit with bhkWorldObject and bhkEntity.

Having each in their own CInfo simplifies both naming and versioning at the cost of repetition.

The bhkRigidBodyCInfo variants will need some kind of shared parent, but compound inheritance is currently not part of the XML spec.  It would have to be done with custom types in niflib.  The shared parent will be needed for getting/setting the same info on different CInfo versions.

Use a correct bitfield for NiGeometryData flags now that we know what the unknown values are for Bethesda bhk data. (NiGeometryData that gets used in a bhk shape)

Specifying ver1/ver2 even for `vercond="#BS...` for FO3 and later allows early rejection of that member for any non-BS 20.2 NIFs.

Unknown Short for FO3+ was two values, a byte and a bool.

Renamed "Num Skin Partition Blocks" and "Skin Partition Blocks" and combined the Partitions array into one row, as having a row separate for SSE was unnecessary.

For #76.  `calc` does not need to be supported and is only used for pre-serialization preparation of the data.  This can also be done by hand.

Some revisions also need to be made for #70 and #73 to account for another attribute with its own expression grammar and tokens.

However, the tokens introduced for `calc` will not be added to the tokens in the XML and will need to be explicitly supported if supporting `calc`.  They are:

1D array size (uses regex): `#LEN[(.*?)]#`
2D array size (uses regex): `#LEN2[(.*?)]#`

Ternary `?`: `#THEN#`
Ternary `:`: `#ELSE#`

Also removes any unnecessary `calculated` without replacement.

Additionally, `calc` is used to limit array size for the time being, though it's possible this should be done with its own attribute, such as `maxlen`.

FO4 DLC uses BSLightingShaderPropertyUShortController in 2 NIFs, but according to the CK it is not actually a functional block and spits out constant warnings. It is still needed in nif.xml for serialization though.

FO4 also adds two booleans for Env Map shader type for SSR.  These were previously marked as an Unknown Short.  These booleans are *not* synced with their values in the BGSM files except for generated meshes such as SCOL and precombined.

Havok serialization is separate from inheritance and so certain shapes that are actually convex were not considered such by nif.xml, which in turn excludes them from uses of `template="bhkConvexShape"`.

Created a fake base class to fix these issues.  Also, bhkConvexTransformShape no longer inheriting bhkTransformShape was more correct anyway, as the Ref type could be made more specific (bhkConvexShape).

Did some comment/description cleanup and some minor decoding.

For #76

For reduced ambiguity.  Adapting for this in Python is as easy as

```py
# XML Booleans
XML_TRUE = ["true", "1"]
XML_FALSE = ["false", "0"]

# Example
self.is_abstract = get('abstract') in XML_TRUE
```

Note "True" and "False" are not valid booleans in XML.  Only lowercase.

Named the compounds correctly after the actual structs,  separated the "Has Data" bool from the compound as it's not actually part of the struct, but before it in a for loop during (de)serialize.

Combined NiAdditionalGeometryData and the BS version into one compound when possible.  But the block can actually no longer be found in any official files, so the BS version is not very important.

Try to use element text whenever possible, and remove any unnecessary XML comments.  Soon the XML will be regenerated and so the XML comments need to be well formed and in predictable locations.

Finally added bhkPlaneShape, used once in SSE.  Added bhkCylinderShape which is possible to export from the official FO4 exporter.

Cleaned up much of the class descriptions and also added `binary="true"` to the padding bytes.

For #3, #76

Replaces all Flags types with real enums (where there were only two options) or a new bitfield type.

Bitfield mimics the bitfield syntax of C structs (Compounds), but also acts like Enum/Bitflags with the `storage` attribute in that the entire structure can simply be aliased to a basic type like `ushort`.

The `type` on each Bitfield member does **NOT denote size**.  Each member only takes up the bits that are defined by `width` and `pos`.  The `type` information is for casting for getters and setters.  That means for C structs, code generation would not use the type on each member, but the storage type instead.

For #70

Their use was discussed as being useful for shifting and masking values before sending as an ARG.

For #74

Implemented thanks to changes done in #76 for differentiating integral and "countable" i.e. can be used for sizes.

Marking types as boolean was also in #76, yet there were no errors for this lint; all the `cond` references that were not complex expressions were already boolean types.