Determine the best approach to derive/downscale the tas variable

[cstephen]

Here's one idea (borrowed from ua-snap/data-api#653):

The downscaling process complicates things a bit since we're using DTR to derive tasmin from tasmax after bias-adjustment, but even while doing so, the per-model relationship between tasmax and tasmin is preserved. My intuition is to use a similar approach to derive per-model tas by calculating the distance between tasmax and tas from the CMIP6 source data, and then subtracting that distance from tasmax after bias-adjustment to derive tas.

We'll need to think this through as a group when it comes time to downscale the next batch of variables.

[Joshdpaul]

Currently, there is no derived mean daily temperature variable produced in the downscaling pipeline; we only have tasmin and tasmax. The climatologies pipeline (in development) derives mean daily temperature on the fly as the simple mean between tasmin and tasmax:

# Derived variables computed inside compute_fragment.py rather than read
# directly from a source family.
derived:
  tmean:
    output_var: tmean
    requires: [tasmax, tasmin]
    units: K
    long_name: daily mean near-surface air temperature
    description: Daily mean temperature, computed as (tmax + tmin) / 2 from daily tmax/tmin before temporal aggregation.

This could potentially be changed to a different calculation if required.