study performance on cosmics-masked coadds

[sbailey]

Study the performance of redrock on cosmics-masked coadds. On a low-statistics cross check of ELG tile 80608 on night 20201216, the results weren't convincing that cosmics masking/coadds were helping.

Comparing redrock fits on spectra (without cosmics masking) vs. desi_coadd_spectra --nsig 4 coadds with cosmics masking, 63 differed by more than 300 km/s. However, only 4 of those were in the truth table /global/cfs/cdirs/desi/sv/vi/TruthTables/Blanc/ELG/desi-vi_ELG_tile80608_nightdeep_merged_all_210208.csv with best_quality>2.5. Of those 4:

TARGETID           Z
39633321168212037  1.0433  orig correct, cosmics-masked coadd changed to 0.78, both with ZWARN=4
39633334912946761  1.6321  orig wrong, cosmics-masked coadd correct, both with ZWARN=4
39633338322913292  1.629   both incorrect, but cosmics-masked coadd incorrectly confident ZWARN=0
39633341661577371  2.371   orig correct, cosmics-masked coadd changed to 1.37, both with ZWARN=4

i.e. it's only 4 targets and all of them had ZWARN=4 in the orig spectra, but 3 out of 4 cases the orig non-cosmics masked version was more correct. There might be some confirmation bias in the truth table since it was also vetted on an earlier non-cosmics masked version of these spectra.