Scaling OLGA and Gliph2 for Compare workflow

[dltamayo]

Running the Compare workflow, particularly OLGA for pgen calculation for tcrsharing and the clustering for Gliph2, does not scale well with increasing cohort size. Currently:

• OLGA calculates generation probability serially
• Gliph2 clustering cpu/memory/time increase exponentially with larger clusters

Possible solutions to implement:
OLGA:

• find means of vectorizing compute_pgen process
• chunk input CDR3 df and parallelize within Nextflow/python
• reformat so that pgen is calculated per sample (ie. take the OLGA outputs from sample workflow and feed into compare workflow)

Gliph2:

• group input by CDR3 across samples to reduce size of input (potentially pivot long per sample afterwards)
• subset Gliph2 inputs by metadata or specific groups of samples so entire cohort isn't clustered at once, and Gliph2 subgroups can be clustered in parallel
  • would enable more direct downstream
  • need to assess biological impact of running Gliph2 on whole cohort vs subsets
• add cluster_min_size as parameter and/or tinker with other Gliph2 args to reduce amount of clustering

[dltamayo]

Implemented in #73