Allow specifying grid acquisition probability model with wildcards

[taldcroft]

The grid probability models are named grid-<name>-<date>. Currently the default model is hardcoded with the full name:

DEFAULT_MODEL = "grid-floor-2020-02"

In order to update the model without a new release of chandra_aca and a new FOT MATLAB tools release, enable wildcards and select the most recent date of matching models. Then use:

DEFAULT_MODEL = "grid-*"

[jeanconn]

Great idea to take steps now to make this more agile. Though could we just chuck DEFAULT_MODEL altogether?

[taldcroft]

DEFAULT_MODEL is very useful in testing because you can change it and then re-run proseco star selection. In other words, anywhere that star probs are used then you can globally override the model without any API changes.

[taldcroft]

With a more modern mindset we could change that module global into an astropy configuration parameter. That's an option which would probably break something but not anything that can't be easily fixed.

[jeanconn]

I was thinking that if you wanted to change the model in testing you'd just control what was in chandra_models or the chandra_models environment variable without API changes.

[taldcroft]

That's quite cumbersome and difficult for the simple case of comparing e.g. the current flight model to a new one. For example:

ef reselect_catalog(aca):
    kwargs = aca.call_args.copy()
    kwargs["include_ids_acq"] = list(aca.acqs['id'])
    kwargs["include_halfws_acq"] = list(aca.acqs['halfw'])
    kwargs["dark_date"] = aca.dark_date
    kwargs["n_acq"] = len(aca.acqs)

    star_probs.DEFAULT_MODEL = 'grid-floor-2020-02'
    aca_orig = get_aca_catalog(**kwargs)
    p2_orig = -aca_orig.acqs.get_log_p_2_or_fewer()

    star_probs.DEFAULT_MODEL = 'grid-local-quadratic-2023-05'
    aca_new = get_aca_catalog(**kwargs)
    p2_new = -aca_new.acqs.get_log_p_2_or_fewer()

    return p2_orig, p2_new

[jeanconn]

Gotcha. I figured that there was already something handy for doing that same thing when comparing different thermal models, but I haven't don't those comparisons myself.