BVHNode improvements

[Walther]

• see https://github.com/Walther/clovers/issues/105
• additionally, think about other possible improvements for BVHNode
• e.g. spatial / octree building instead of choosing a random axis per node split
• e.g. consider other structures than binary tree; e.g. B-tree, Finger tree, others
• inspect the built tree for possible pathological cases, e.g. get debug printouts of how the tree got built
• think about what a "sensible" tree would look like: balanced? unbalanced? spatially correlated? dense near objects?
• other possible replacements, e.g. bounding interval hierarchy ?

[Walther]

BVHNode splitting done by longest axis instead of random axis done in #140

[Walther]

Alternative: https://pbr-book.org/3ed-2018/Primitives_and_Intersection_Acceleration/Kd-Tree_Accelerator

[Walther]

Look into modern papers and blogs!

• Fast Parallel Construction of High-Quality Bounding Volume Hierarchies
• How to build a BVH

Add statistics methods for inspecting the "quality" of the constructed BVH.

[Walther]

First set of improvements done in #208. Closing as complete - future improvements can be done as separate tickets.