Fix TSNR value when cframe mask or ivar=0

[julienguy]

This PR fixes 2 issues in TSNR calculation:

• avoid situations where we apply a correction based on a ratio of 2 small numbers for fibers that are off target (and have a low TSNR anyway).
• use the cframe.mask==0 and cframe.ivar>0 instead of frame.mask==0 and frame.ivar>0 in the computation of TSNR because some spectal values are masked during calibration.

[julienguy]

Validation example:

desi_process_exposure --infile /global/cfs/cdirs/desi/spectro/redux/jura/exposures/20231206/00208376/frame-r2-00208376.fits.gz --fiberflat /global/cfs/cdirs/desi/spectro/redux/jura/exposures/20231206/00208376/fiberflatexp-r2-00208376.fits.gz --sky /global/cfs/cdirs/desi/spectro/redux/jura/exposures/20231206/00208376/sky-r2-00208376.fits.gz --calib /global/cfs/cdirs/desi/spectro/redux/jura/exposures/20231206/00208376/fluxcalib-r2-00208376.fits.gz --outfile cframe-r2-00208376-new.fits.gz --cosmics-nsig 6

Two fibers that had large TSNR now have TSNR values=0 (because masked), the others are unchanged.

[sbailey]

Thanks. I double checked this case for main vs. this PR and confirm that 2 fibers that are very offset and completely masked now have TSNR2=0 instead of a crazy large value, and 4 other fibers have a very small decrease in TSNR2 which we interpret as being because cframes have slightly more masked pixels than frames (e.g. due to fiberflats or fluxcalib having masked pixels).