Apply per-direction weights in constraints

This feature makes sure that constraints can use direction-dependent weights derived from the model data. This means that data with low or zero model data will have less/no effect on the constraint. This is particularly useful for getting smooth solutions of e.g. a-team sources that might not have sufficient S/N ratio over the full solved spectrum. In that cases, areas of low S/N (e.g. nulls) may have bad solutions and, if not downweighted, skew the smoothed values.

I tested this on a LOFAR data set that we use for Rapthor, which includes a smoothness constraint, and I see small changes in the solutions. I think that confirms it a) does something; and b) does not do something crazy. Stefanie of the LOFAR EOR team will test the feature further.