Bug fix/conditionally compute tau cirrus after met downselection

[tsankar]

Changes

• Adds "compute_tau_cirrus_in_model_init" parameter which determines when tau_cirrus is computed.
• In "auto" mode, computes tau_cirrus in init iff met is dask-backed. Otherwise, computes it after downselection.

Breaking changes

None

Fixes

• Memory explosion when using CoCiP on non-dask-backed weather data.

Tests

• QC test passes locally (make test)
  -- GCP tests in test_cache.py fail locally, presumably because of auth issues?
• CI tests pass
  -- Let me know the best way to run these if necessary!

Reviewer

@zebengberg

[zebengberg]

Thank you Tharun, this is great!

I made some a commit straight to main while your PR was opened ... can you rebase off main then re-push?

[zebengberg]

Does it make sense to remove the default values for compute_tau_cirrus (and for shift_radiation_time)? If we only ever pass model parameters into this function, then we may be less likely to create future bugs if these parameters don't take defaults. But if we're calling process_met_datasets in places where we don't specify compute_tau_cirrus or shift_radiation_time, leave as is.

[tsankar]

The only other place we call process_met_datasets is in tests/benchmark/north-atlantic-study/validate.py, is this benchmark test still used? I agree that it would be safer to not include defaults here.

[zebengberg]

We've been meaning to run the benchmark tests occasionally in the github CI. That hasn't happened yet, and so I think there are many slightly broken things in the benchmark directory at this point. No need to worry about that!