QA with Truth (S/N and redshift refactor)

[profxj]

New code for generating QA (plots) using the Truth.
Two areas are S/N plots (as a function of object type)
and refactoring redshifts.

Here are the main changes:

• Add desi_qa_s2n script
• Refactor desi_qa_zfind script
• Refactor redshift QA for new data model
• Refactor shared QA methods to desisim.spec_qa.utils
• New plots for S/N of spectra for various objects (ELG, LRG, QSO)

Here are some examples of the S/N and new redshift plots:

[sbailey]

Looks good. We'll have some future tweaks but can merge now to get the updated data model structure in place.

I'm curious why the faintest LRG bin had higher S/N than the second faintest bin. Low statistics fluctuation? Double observation of the faintest and single observation of the sort-of-faint ones? We should adjust the mag bin boundaries to match the 1-,2-,3-observation LRG (soon to be split out into separate targeting bits). Currently the boundaries are

• zmag < 20.36 : 1 obs
• 20.36 <= zmag < 20.56 : 2 obs
• 20.56 <= zmag : 3 obs

I'm sure those numbers will be adjusted in the future and we shouldn't arbitrarily hard code them here, so this is mainly a comment as a heads up for future consideration.