This drastically reduces the number of allocated float arrays on the heap just by flattening the uvs into a single array per model