Better bookkeeping for deep images

[AHorneffer]

The deep-imaging Initial-Subtract pipeline (discussed in #7) should offer more flexibility for the imaging:

• Image not all bands but just a given number of bands together, so that there are more imager processes running at the same time. (E.g. specify num_bands_per_image)
• Do not generate one channel-image per band but only one per a given number of bands and then use the same sky-model for a number of bands. (E.g. specify num_bands_per_channel)

That way by setting num_bands_per_channel = 1 and num_bands_per_image = 1 one gets the behaviour of the "old" non-deep initsubtract, and by setting num_bands_per_image = num_bands the behaviour of the deep init-subtract, and with other settings one can fine-tune the performance.

When this is done we only need one Initial-Subtract pipeline anymore.

[twshimwell]

In this updated script it would also be good to adjust the threshold to a suitable value that depends on the amount of data being imaged. The threshold as that can be estimated from the amount of data (which is already known in magic script) as say 5x0.1/(bands^0.5*inttime^0.5) which corresponds to 35mJy for an 8 hr observation with 24 bands. This is most likely an overestimate of five times the noise but I think would probably be ok for the high1 and low1 images.