Feature/chunked writer precomputed burdens testing combo

[bfclarke]

What

This is a fairly large PR that makes several changes:

1. Splits the computation of burdens from the formatting of phenotypes and covariates (this makes a cleaner pipeline, but is especially important when using precomputed burdens to do association testing, for which a pipeline has been added).
2. Fixes a bug involving multiple concurrent writers to a Zarr array, where burdens were sometimes not written.
3. Adds a deterministic flag to the config - useful when testing results of pipeline runs.
4. Reduces the number of phenotypes used for training and association testing in the example config.
5. Adds an example input config for the REGENIE pipelines.
6. Adds a pipeline for conditional association testing with REGENIE.
7. Moves the logic for when to (re)generate configuration files to config.py
8. Adds a script for comparing pipeline results to reference results
9. Adds some additional tests

Testing

• The CV training/association pipeline has been run on example data and the results checked against those from the current main branch
• A real-data check has been run on UKB WES

[HolEv]

spottet only some minor things we could still fix for consistency.
Main thing is that we have two log folder is burdens log and logs which we should unify. Otherwise really nice!