I made the BSP generation much faster by in-lining subroutines and unrolling small loops, but it reduced the amount of nodes in the tree (probably because I removed the proper calculation for the greatest dot product). I also slightly improved the traversal algorithm (removing ~1 arithmetic operation per function call).
Credits: @tsf70 for the original project Speed: Sub 0.5 second level build time on a Chromebook, with both 'bounds' settings.
