[JeffM2501]

This PR adds a simple optimization to GetRayCollisionMesh that in my testing improves the raycast speed by about 50% in the mesh picking demos.

The current code transforms every vertex in the mesh into worldspace to compare it against the worldspace ray.
This means that we have to do 3 transforms per triangle, and that adds up, fast.

This PR instead transforms the ray into model space so that we can just test the vertex data without doing any math on them. The point is transformed by the inverse of the transform matrix once, and the direction by the rotation of that matrix, once.
This is much faster than transforming every vertex for every triangle.
There is a fix up step at the end that transforms the model space hit pos and normal, back into worldspace, but again this is one transform total.

With this change, we do a max of 4 transforms per raycast check, not Triangle count * 3.

[raysan5]

@JeffM2501 thanks for the improvement!