Add support for non-linear darks to dark_night

[sbailey]

We now generate a nightly linear dark using desi_compute_dark_night. However, at the start of DESI we needed to use non-linear darks generated by desi_compute_dark_nonlinear. AFAIK neither the pipeline wrapper nor desi_compute_dark_night knows when or how to fallback to using a non-linear dark when needed. To make matters worse, I don't think we have any automated way to know if a non-linear dark was needed, so we might just need to hardcode 30 dates before which non-linear darks are required.

To do:

• determine those dates when we can start using linear darks
• update the nightly dark logic in the pipeline so that future prods will know whether to make a non-linear or a linear dark night

This is a blocking factor for Matterhorn, and thus high priority.

@jbanovetz please take the lead on this, consulting with @akremin and @julienguy as needed.

[jbanovetz]

After running into some issues with corner cases, I have confirmation that my change works! This is with branch darknight_date_cutoff and it adds in 2-3 lines of code that introduces a cutoff date of 20240509 as past this date, there is no more non-linearity in the darks with any of the cameras (see https://desi.lbl.gov/DocDB/cgi-bin/private/ShowDocument?docid=9765 for more details). These changes are to only make darknight models after the cutoff date and to start preprocessing darks 20240408 (30 days before the cutoff). So non-linear darks will be used before the cutoff date and linear darks (via darkinght) will be used afterwards.

The tests that I ran:
Using dry runs to just produce the jobs
Date after 20240509 produce a darknight and preprocesses darks
Date before 20240509 and after 30 days previously doesn't produce a darknight and doesn't make preprocessed darks
Date before 20240509 but within 30 days doesn't produce a dark night but does make preprocessed darks

Using full runs:
Date before 20240509 and after 30 days before used the DESI_SPECTRO_DARK dark to process the images
Date after 20240509 used the darknight to process the images

If these tests are sufficient, I will make a PR with the changes.

[akremin]

I think these tests are sufficient. Please proceed with a PR, and we can take a look.