Explore removing array -> box conversions

[gkjohnson]

Related to #656

• Adjusts shapecast to pass a buffer rather than Box3 with an option to use Box3

Remaining Issues

• Make a bvhcast and intersects geometry PR separate from "shapecast"
  • Test performance on bvhcast, intersects geometry
• Shapecast needs to provide unique instances of box and the bounds array to ensure recursive calls work correctly
  • Use a stack / pool
• Test shapecast perf benefits first.

[gkjohnson]

cc @agargaro - awhile ago we discussed improving the performance of the "shapecast" function by allowing it to pass float32 "bounds array" rather than implicitly converting it a Box3 ahead of time. This PR makes (most of) the changes to achieve this with the "useBox3 = false" option. ClosestPointToPoint has been changed to use the bounds arrays but I'm not seeing any performance benefit compared to master. Is this along the lines of what you were thinking?

[agargaro]

I would have expected a slight improvement, like with raycasting. I'll try running some tests locally later to understand more 😄