Add calls to UnloadFileData() before return in cases of invalid IQM file.

* remove duplicate calculation of `invRotation` in `UpdateModelAnimationBones`

* fix normal transform in `UpdateModelAnimation`

* Update rmodels.c

resolves segfault with missing bone weights or bone IDs

* Update rmodels.c segfault with animation and missing normals/animnormals

* correct place.

- Updating bones only once instead for each mesh.
 - Updating only one `model.meshes[].boneMatrices` and then using deep copy for other meshes instead of calculating for each bone in each mesh.

**Other points:**
 - Makes it a clean base/template/reference for bone updation functions. Because if using this as template then some calculations done in one mesh can affect bones in other mesh in next iteration(doubles the effect in for next mesh).

Signed-off-by: Kirandeep-Singh-Khehra <kirandeepsinghkhehra@gmail.com>

* [rmodels] Fixed null pointer dereference in LoadGLTF

* [rmodels] Add parenthesis around conditionals in LoadGLTF

* increased vertex data precision to match blender exports

* modified memory allocation to better fit higher precision data

* now accounting for newline characters during allocation

* now accounting for potentially higher vertex counts

* removed unnecessary final newline

The nextShapeEnd integer is a pointer in the OBJ data structures.
The faceVertIndex is a vertex index counter for the mesh we are
about to create. Both integers are not compatible, which causes
the code to finish the meshes too early, thus writing the OBJ data
incompletely into the target meshes.

It wasn't noticed because for the last mesh, it process all remaining
data, causing the last mesh to contain all remaining triangles.

This would have been noticed if the OBJ meshes used different textures
for each mesh.

If we borrow from the GPU skinned animation code, we can just multiply the vertex by the matrix * weight and shave a chunk of CPU time.

Still, not fully convinced of those functions naming, despite quite descriptive, sounds a bit confusing to me...

When creating a new sphere mesh with high number of slices/rings the top and bottom parts of the generated sphere are removed. This happens because the triangles in those parts, due to high resolution, end up being very small and are removed as part of the 'par_shapes' library's optimization. Adding par_shapes_set_epsilon_degenerate_sphere(0.0); before generating the sphere mesh sets the threshold for removal of small triangles is removed and the sphere is returned to raylib correctly.

The indication of locations for bone ids and bone weights did not correspond to their default values ​​in config.h

Only warns when there are more animations than currently implemented
Allows mesh indices to be unsigned char

* Fix rlgl standalone defaults

* Fix rmodels

* Make the max VBO match the animation flag.

* re-enable GPU skinning for mac, and fix the max buffer to be correct based on the GPU skinning support flag.

* Update raylib_api.* by CI

* Disable GPU skinning on MacOS
Add GPU skinning example to MSVC Projects.

* Update raylib_api.* by CI

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

* Added optional GPU skinning

* Added skinned bone matrices support for different file formats.

* Moved new shader locations to end of enum to avoid breaking existing examples. Added gpu skinning on drawing of instanced meshes.

* Added GPU skinning example.

* Removed variable declaration to avoid shadowing warning.