niflib: removed -fpermissive flag from Makefile and fixed 32 bit pointer casts for 64 bit build (reported by guidoj, fixed with hints from Alphax, see issue #2788607)

   - Fix issue with exporting Skin data causing missing vertexes in nifskope
   - add helper methods for collision related export.
   - fix crashes when exporting animation

Finished new MatTexCollection helper class which makes it possible to manipulate material properties and texture objects with the same classes regardless of the intended NIF version.
Realized that only read was implemented for files with version < 3.3.0.13, so implemented write for those old versions as well.

Since there seemed to be no way to make pre-main initialization work for statically linking the library, I went back to using a CPP file that includes all the NiObject files and registers each one with the ObjectRegistry.

Moved Object Registration/Creation functionality into new ObjectRegistry class.  Hopefully this fixes the issue with the object map not being initialized yet since it is a static member of the class and should theoretically be initialized before code in functions of that class can be executed.  Seems to work for me, but I'll wait until Tazpn tests it to declare it fixed.

- There are no longer any defines, things like inheritance class name, NIF attribute variables, and default constructor code are all visible without looking at another file.
- All non-custom parts of the Object files are now totally identical and in the same order, so it will be easier to compare them.
- NiObjects are now completely self-contained.  To remove them from Niflib, you have only to exclude their CPP file from the build.  To add new ones, you only have to include their new CPP file.
- Making a change to a NiObject implementation file only causes that one file to be re-compiled, which is generally much quicker than before since the large obj_define and obj_impl files do not have to be parsed or compiled.
- There are no longer any "Internal" versions of the Read, Write, asString, FixLinks, or GetRefs functions.  Now there is just one function for each of these with pre and post custom code areas (except for GetRefs which I didn't think needed it.)
- Updates to the hierarchy in the XML should now be almost completely automatic.  All that should be necessary to make an update is to add any new NiObject CPP files to the build.
- Changes to the generated parts of the NiObject files can now be made at any time, and immediately applied to all NiObjects without editing each existing one.

Updated Niflib to the latest XML which adds support for bhkMeshShape, NiMultiTextureProperty, NiRawImageData, and NiSequence.

Objects now have a Create methods and register themselves in the global object map with pre-main initialization, making obj_factories.cpp obsolete.

Re-exposed TYPE constants and marked them NIFLIB_API, making the TypeConst function obsolete.  You now must use NiObject::TYPE instead of NiObject::TypeConst() again.

Added ClearHardareSkinInfo function to NiTriBasedGeom so that SetSkinPartition functions could be made NIFLIB_HIDDEN.

Laying the groundwork for Niflib to be able to work on systems with different endian types.  All Read/Write functions now take NifInfo structures which are based by const reference.  These are also passed into the NifStream functions so they will have access to the requested endian type of the NIF as well.  Also added endian detection and swapping functions.

Moved TypeConst function implementation out of header file. Hopefully should fix GCC visibility issue.
Fixed GetIDString virtual function missmatch that was causing the name not to show up when the function was called from a NiObjectRef.
Added Python-specific operator functions to Ref to prevent the wrapper from crashing on Null Refs.

Properly capitalized include file names in niflib.cpp.

Fixed a bug in ComplexShape::Merge that occurred when a shape had no UV set, but, for some reason, had a texture referencing UV set zero.
Changed NiSkinData::NormalizeWeights to use doubles, trying to increase the accuracy.

ComplexShape::Split now cuts out weights below 0.1 and normalizes the result.
Fixed a bug which could cause NiSkinInstance to try to call a function on a NiNode that was in the midst of destruction.
Removed some cout statements.

BindSkin no longer automatically clears transforms between the mesh and the root. Instead this can be done with the ApplySkinOffset function after all the shapes are in place.

Corrected Shape/particle hierarchy by adding NiGeometry and NiGeometryData.  This is the correct hierarchy and having particles come back as derived from NiTriBasedGeomData was causing problems.
Commented out printf commands in triangle stripper.

Replaced _WRITE, _READ, _STRING, AND _FIXLINKS defines with the corresponding internal function call.
Replaced STANDARD_INTERNAL_FUNCTIONS with the internal function signatures.
Made my Doxygen config file use relative paths and added it to SVN so everyone can generate the same documentation for themselves.
Removed includes WZ added to ComplexShape.  These were not missing, they are up to the user to include if they wish.

Added support for sub version 4 headers, object storage, and link method.  Some objects like NiNodes already read correctly.

"Fixed" NiTriBasedGeom::GetSkinInfluencedVertices function so it fails to import meshes with bad overall offset transforms.

Really put the skin calculations back the way they were this time.  Helps, but the hands of better bodies meshes still end up in the wrong place.
Added transform parameter to ComplexShape::Split because changing the transforms after the fact ruins the skin binds.

Fixed the relative include paths and tried to update solution, but havn't got the precompiled header working right.

Initial changes for skin partition access.  Needs more work around protecting the user from inconsistencies.