Downsampling routines + tests

[Algomorph]

Necessary for proper node topological updates, it is imperative we get this right.

• Finish elementary radius-based (epsilon) downsampling (based on existing grid downsampling code)

Perhaps make a version that exploits spatial hash map instead of a linear bin array. This will involve:

• Write a routine that determines how many maximum points we might have per "epsilon-radius" block
• Write a kernel routine for aggregating points into hash blocks
• Write a kernel routine for averaging points within each block
• Write a kernel routine for checking block neighbors and merging neighboring blocks' points if they are within epsilon