Region processing speedup

[thekugelmeister]

Quick fixes to speed up regions_to_list, by omitting parallelization infrastructure when running with a single core.

Updates at a glance

• split *_process_chunk functions into a processing function and a higher-level parallelization function, for both pileup counts and vectors.
• updated the user-facing counts and vector functions to skip initializing worker pools and shared memory when running with a single core
• For running regions_to_list on pileup_counts_from_bedmethyl, this results in ~300x speedup

Other findings

While implementing these fixes, a few things have become clear through testing:

• Initializing worker pools and shared memory has significant overhead in the current implementation
• Reopening tabix files with pysam has significant overhead

Below is a plot comparing processing speeds for a set of simplified tests (blue) and different versions of pileup counting methods (orange). In addition to identifying the main cause of the regions_to_list slowdown, this visualization suggests that there may be a ~2x speedup available if we avoid reopening tabix files as much as possible.

This change will be implemented as part of a future effort.