Can we vectorize some of compress by redundancy?

Noted in #827 this block of code is used in the averaging feature of the compress by redundancy, but it is possible we could replace the for loops with a well constructed search much like the uvw calculation. This would require a baseline,time ordering to be enforced for this option though. Investigation is needed.

                for bl in reg_group:
                    bl_inds = np.where(self.baseline_array == bl)[0]
                    group_inds.extend(bl_inds)
                    group_times.extend(self.time_array[bl_inds])
                for bl in conj_group:
                    bl_inds = np.where(self.baseline_array == bl)[0]
                    conj_group_inds.extend(bl_inds)
                    conj_group_times.extend(self.time_array[bl_inds])

Start in the averaging implementation for compress by redundancy and compare its baseline and time handling with the uvw calculation mentioned in the issue. Investigate whether enforcing baseline,time ordering enables a search-based replacement for the shown loops. Done means the averaging path uses the investigated vectorized approach and its behavior remains correct.