Task #11550: Don't output maxiter when stopped early, don't calculate stepsize over failed solutions as it would always become NaN

Task #11550: Fixing wrong indexing of solutions, causing part of solutions not to be smoothed when ndir>1

Task #11550: Further improved the convergence criterion. It is now (finally) completely scale independent and uses a fully squared metric to upweight larger errors somewhat.

Task #11550: Comparison to see whether tolerance has been multiplied by stepsize^2, correcting default precision level to accomodate this change

Task #11550: Fixing various cases where a zero-weighted antenna could cause infs in intermediate values (tec / solutions), leading to too many iterations

Task #11550: Fixing how weights are handled by TEC constraint: old code would use antenna-averaged weights per antenna, and failed matrix inversions in the solver would cause random channels of other antennas to be flagged