Optimize face ordering for post transform vertex cache

[ArcRiley]

To date we've only done the most basic optimization of face ordering; grouping faces by material to minimize state changes, then rendering a subrange of the VBO element array for each material.

For smaller meshes (all we've had so far) this is fine, but larger meshes this can result in recalculating the same vertices repeatedly each time a mesh is rendered. To minimize this we can sort faces (within each material range) by their vertices. OpenGL hardware will then be able to use the post transform vertex cache data which can make a difference on larger meshes or smaller hardware.

This will become important the more we use vertex shaders for mesh animation.

 http://www.opengl.org/wiki/Post_Transform_Cache