Don't use mask information to mark bad pixels in SDSS.

[weaverba137]

This PR closes #92.

As a stretch goal for this PR: are there ever cases where we want to mask DESI pixels above and beyond cases were inverse variance == 0? Are there some cases where a mask value should be ignored, versus "definitely do not use this pixel, even if inverse variance != 0", or somewhere in between?

[coveralls]

coverage: 2.032% (-0.004%) from 2.036%
when pulling 338e80c on sdss-masking
into 9710948 on main.

[sbailey]

I believe for DESI, all specmask bits are treated as bad and already have ivar=0 set too. I haven't traced it back in the code where this might be enforced, but that seems to be true from spot checking coadds. None-the-less, Redrock also specifically sets ivar=0 for any mask!=0 (redrock/external/desi.py L428), which might just be a belt-and-suspenders extra step.

For context, this is different for the fiberstatus mask, which has informative but non-fatal bits like RESTRICTED and POORPOSITION.

[weaverba137]

@sbailey, thank you. I will not make any changes to DESI masking for the purposes of this PR, but I will add a condensed version of your reply in a comment.