No way to reject outlierss in doing overscan fit

As reported by @tapio-not:

I encountered a new issue with the astropy/ccdproc, specifically
`subtract_overscan`.

The "problem" is that the data I have has occasional bad first line
with values of 4e9 instead of about 10 000 (ADU). This is a CCD-controller
issue, so I just have to live with that.

So, if I fit " models.Polynomial1D(1)" this fails because I have not found
"sigma clipping" option in this "task" as with iraf/ccdproc there is
"Low and high sigma rejection factors for rejecting deviant
points from the overscan fit."

Of course I can replace the 4e9 values with something smaller, but much
more elegant solution would be to have "sigma clipping" while fitting the
data. I guess that using "NDData uncertainty" could do the job,
but I'm not familiar enough with that and anyway this might be an
overkill in comparison to reject one/few points from a fit.

Start with the ccdproc.subtract_overscan entry point and its Polynomial1D fitting path, then check how NDData uncertainty is currently handled. Done means overscan fitting can reject one or more deviant points through sigma-clipping behavior, without requiring callers to replace the bad values manually.

Written by the indexing model from the issue text.