[FEATURE] Carefully decide what images get used for WINTER flats

[virajkaram]

Is your feature request related to a problem? Please describe.
WINTER flat-fielding happens with images that don't necessarily have a large numbers of counts. Looks like the dark counts vary on a level of few hundreds. If the sky counts in the flats are ~2000, this means that our winter frames would never be calibrated better than 10% (min. uncertainty is 0.1 mag, before including any other effects). This is very suboptimal.

Describe the solution you'd like
We should set a threshold on the median number of counts in an image that should be used for flat-fielding, to improve the quality of reductions. This would probably translate to only using frames taken earlier during the evening. On an astro-note, we should probably be taking twilight flats.

[robertdstein]

I think this has been somewhat superseded by dome flats. See #835 instead.