Performance: Combining flats is slow

This is taken from Astropy project performance survey. Is this is not the correct repo for this issue, please advise.

I have not tried this in recent astropy versions, but the early code to combine flats was FAR slower than IRAF `imcombine`. I had done some reading up to discover that IRAF was very proud about how it did memory management to do this efficiently. Sorry cannot be more specific or useful than that.

The report names no implementation file, test, benchmark, or recent Astropy version. Start by locating the current flat-combination path in ccdproc and reproducing its performance with a representative dataset, then compare the result with IRAF imcombine. Done would require a defined benchmark and an agreed performance target, neither of which the issue currently specifies.

Written by the indexing model from the issue text.