fix FIRSTNIGHT/LASTNIGHT when missing petals

[sbailey]

This PR moves the FIRSTNIGHT,LASTNIGHT calculation from desispec.coaddition.coadd_fibermap to desi_zcatalog (the script the stacks N>>1 redrock files into the final redshift catalogs) to fix a problem on tiles that are missing different petals on different nights. Although coadd_fibermap was correctly considering all nights in its input fibermap, the problem is that those per-petal input fibermaps can have a different set of nights per petal, resulting in different FIRSTNIGHT,LASTNIGHT values per-petal when they are supposed to be per-tile values. This is especially important for LASTNIGHT which is used to trace back to the original spectra in tiles/cumulative/TILEID/LASTNIGHT/spectra-PETAL-*.fits.gz (same LASTNIGHT for all PETALs, even if some PETAL wasn't included on that LASTNIGHT)

For example, tile 80730 was missing petal 3 on the first night it was observed (20210218), and missing petal 6 on the last night it was observed (20210322), resulting in a mix of FIRSTNIGHT, LASTNIGHT values different per petal.

Example outputs are in $CFS/desi/users/sjbailey/dev/coadd_lastnight:

• spectra*.fits.gz - original spectra from iron
• coadd-*-iron.fits - original coadded spectra from iron
• coadd-*-main.fits - coadded spectra from current main
• coadd-*-pr.fits - coadded spectra from this PR
• zcat-80730-iron.fits - stacked redshift catalog using iron-era code (release 23.1, desispec/0.56.5)
• zcat-80730-main.fits - stacked redshift catalog from main
• zcat-80730-pr.fits - stacked redshift catalog from this PR

Issues / differences to note:

• coadd-*-iron.fits FIBERMAP does not have FIRSTNIGHT/LASTNIGHT
• coadd-*-main.fits FIBERMAP has FIRSTNIGHT/LASTNIGHT, but they are different
  for each file
• coadd-*-pr.fits FIBERMAP does not have FIRSTNIGHT/LASTNIGHT
  (like the original iron coadded FIBERMAPs)
• zcat-80730-iron.fits no FIRSTNIGHT, has single LASTNIGHT=20210322
• zcat-80730-main.fits multiple values of both FIRSTNIGHT and LASTNIGHT
• zcat-80730-pr.fits FIRSTNIGHT=20210218 and LASTNIGHT=20210322, matching
  the first and last nights the tile was observed on any petal.

There were other changes to consolidate the MIN/MAX/MEAN_MJD calculations
(which remain per-target, not per-petal nor per-tile); remove no-longer necessary special casing of RA/DEC/MJD; and remove comments pointing out problems that have been fixed.

Code snippets used to generate example files

for PETAL in 3 4 6; do
  desi_coadd_spectra --onetile \
    -i spectra-$PETAL-80730-thru20210322.fits.gz \
    -o coadd-$PETAL-80730-thru20210322-main.fits
done

desi_zcatalog --recoadd-fibermap --nproc 10 --group cumulative \
  -i $CFS/desi/spectro/redux/iron/tiles/cumulative/80730/20210322 \
  -o zcat-80730-main.fits

Python snippets to evaluate files

Current main has different values of FIRSTNIGHT, LASTNIGHT per petal:

for petal in (3,4,6):
    fm = Table.read(f'coadd-{petal}-80730-thru20210322-main.fits', 'FIBERMAP')
    firstnights = set(fm['FIRSTNIGHT'])
    lastnights = set(fm['LASTNIGHT'])
    print(f'{petal=} {firstnights=} {lastnights=}')

-->
petal=3 firstnights={20210221} lastnights={20210322}
petal=4 firstnights={20210218} lastnights={20210322}
petal=6 firstnights={20210218} lastnights={20210224}

This PR fixes LASTNIGHT to a single value matching iron, while also having FIRSTNIGHT=earliest night of any petal

for version in ('iron', 'main', 'pr'):
    zcat = Table.read(f'zcat-80730-{version}.fits', 'ZCATALOG')
    for col in ('FIRSTNIGHT', 'LASTNIGHT'):
        if col in zcat.colnames:
            tmp = np.unique(np.array(zcat[col]))
            print(f'{version:4s} - {col}={tmp}')
        else:
            print(f'{version:4s} - no {col}')

-->
iron - no FIRSTNIGHT
iron - LASTNIGHT=[20210322]
main - FIRSTNIGHT=[20210218 20210221]
main - LASTNIGHT=[20210224 20210322]
pr   - FIRSTNIGHT=[20210218]
pr   - LASTNIGHT=[20210322]

@akremin please review; @stephjuneau heads up if you are available too.

This is a blocking factor for regenerating patched DR1 redshift catalogs, but the issues are subtle enough that I would appreciate a careful independent review. Thanks.

[akremin]

The code changes look good and the validations seem thorough to me. I'll approve, but won't merge until @stephjuneau has had a chance to comment.

[sbailey]

I promised @stephjuneau that I wouldn't put her on the critical path for any DR1 stuff while she's busy with other topics, so I'm going to go ahead and merge this so that I can make updated iron catalogs, and any post-facto comments can be handled in other PRs.