Generalize Mark 4 decoding

[mhvk]

#332 made clear that there are Mark 4 data in the wild that do not use the default track assignments. Closer reading of the docs indeed makes clear that those defaults were only recommendations for VLBA format, not fixed.

In principle, the Mark 4 header encodes all the information needed to determine how to decode the data, but will need a bit more information that just the (nchan, nbit, fanout) used now; instead it should use the full magnitude_bit array and probably the channel and sideband information. It should not even be too hard to auto-generated decoders/encoders on the fly. If that is done on first encounter, with the encoders/decoders dicts updated, it would not slow down things too much either.