Explore parallelization of ancestry_analysis

[smlmbrt]

Some places where additional cores may get a speed increase:

• n_jobs in random forest classifier https://github.com/PGScatalog/pgscatalog_utils/blob/cf5d0d58877700536c315d645a6810bcdcf44b14/pgscatalog_utils/ancestry/tools.py#L160
• In pgs_adjust writing the adjustments as a loop (this will likely depend on how costly the repeated subsetting is):
  1. Subsetting data
  2. fitting adjustments
  3. return adjustment

[smlmbrt]

I think the speed problems are actually solved by this: https://github.com/PGScatalog/pgscatalog_utils/pull/81/commits/a71344351205d93c7e8d6273e3541040773f5af0