Please add the ability to process n > 1 visits together for cosmic ray rejection

Now that we have Gen3 we can start to add scatter/gather steps to the pipeline, and the first two that come to mind are scattered light and cosmic ray rejection.   This ticket deals with the latter.

Cosmic ray rejection is hard, and we have not managed to tune the single-visit CR code to remove 100% of contaminated pixels, while preserving 100% of uncontaminated pixels.  The normal way to handle this is to compare two or more exposures, and while this isn't totally trivial with PFS (and it's harder for two then for three or more).  I have written code, in a notebook, which appears to do a very good - but not perfect - job; it is, however, very much better than the current approach and I think we should implement it.

CR rejection is important, as it can degrade the sky subtraction as well as generating spurious emission lines.