Ensure that the fallback to pass through long baseline antennas with too much flagging works with the updates to have per-SPW caltables

[psheehan]

I believe, though it has now been some time so my memory is a little fuzzy, that this initially came about because once the per-SPW/BB caltables were added, there was a case where the initial attempt at calibration picked either per-SPW or per-BB as the preferred mode. That solution failed, so it did the combinespw fallback and still failed because of a beam-size change. It thus did the "calonly" fallback, but because the combinespw fallback had failed, per-SPW and per-BB were still options. I don't recall which mode it ultimately picked, but regardless, my recollection is that the selfcal actually continued successfully, and it didn't break until the weblog. At that point, because unflagging had been done, when the weblog looped over all modes for that solint, it was looking for the "*.pre-pass" version of the gaintable, but could only find it for the preferred mode and therefore crashed.

After some thought, I decided that the simplest solution would be that in the event that we reached the "calonly" fallback, that only the combinespw option should be considered, and all other modes should be removed for future solints (resetting for ap solints).

It also, as I was working on this, occurred to me that the option for antennas to fall back to previous successful solint solutions rather than being "passed through" should only consider previous combinespw solints, as others all get applied. It would also make for very confusing code/gaintables to be mixing per-spw and not...

This PR fixes both of those issues.

[psheehan]

It occurred to me, as I was creating this PR, that we used to allow the inf_EB to do a per-SPW unflagging, and this update kills that. I will try to fix that now.