When "bad" flats are used (presumably ones with negative average flux, but possibly other things), the constructFlat.py script fails (in that case due to log scaling giving nans).

In order to look at the underlying cause it would be helpful for it to announce which ccds/visits contained the data that was making it fail.