Update bvhcast

[gkjohnson]

Fix #257

Add new bvhcast implementation

TODO

• Test performance (it's slower)
• Get the number of triangle tests to be identical
• Get the number of matrix applications to be lower / the same
• Can we effectively reproduce what's happening in the original shapecast approach but with some optimizations if we perform it manually?
• GC improvement
• More performance improvements, optimization
• Get the number of bounds traversed lower
• Check bounds in more places
• Ensure correctness (drag position all the way up, revert to prior commits) (broke in a31cb89)