CAS-14874 : Fix for nproc dependent beam size
The reported issue on the ticket was caused by an incorrect cache
refresh of the number of channels in an SPW.
If there are N input MSs, if the N-1 SPW has more channels than the Nth
SPW for the same SPW row, the delta in the number of channels is filled
with zeros that leads to an incorrect Briggs weighting calculation.
The fix was to correctly update the SPW channel count per MS, regardless
of which MS we are currently processing internally. Previously this
would only affect the final MS in a list.
After the fix, the example posted to the ticket passes successfuly, with
all tested parallel breadths yielding the same result.
