Improve FBX Import to buffer algorithm

[athanggupte]

The current implementation of the FBX Importer in backup-changes-290722 branch creates a separate vertex for each triangle thus making the index buffer obsolete as there is no reusing of vertices. This also means that the size of the meshes in memory is larger than required. Fix the algorithm to first extract arrays of all the components - Position, Normal, UV, etc. and then use the metadata in the FBX scene to map them together in a way that uses the memory optimally and also maintains proper hard and soft edges.