Error handling instead of assertions

[finnbear]

#115 (comment): Removing the assert functions from the API. They should likely be only available for testing I suppose. Wouldn't really want to offer functions that crash the program knowingly. We should convert those to proper errors. Maybe use thiserror to ease this a bit.

My thoughts:

• When possible, inputs should be checked (at least debug_assert! but an error may be preferable), instead of creating an invalid BVH or ray.
  • Decide whether AABB's must have volume. If so, ensure they do.
  • Ensuring numbers are within valid ranges (especially not NaN)
• Panics should be avoided
• Essential panics should be documented
• As an additional debugging tool, there could be a way to export BVH AABB's and ray(s) as a 3D model (.PLY seems easiest), which could be examined in 3D software.