Addition of direction dependent simulation

[JSKenyon]

This is very much a pipe-dream. Currently, I compute DD visibilities by calling the solver multiple times with different sources on each call. This works but I suspect it is inefficient, especially when it comes to the computation of parallactic angles etc.

Ideally, it would be awesome if the interface to MB could support this in a more organic fashion.

[sjperkins]

@JSKenyon Would visibilities form multiple fields/directions be summed together or should they be separately calculated?

[JSKenyon]

@sjperkins They would be separate. The goal would be to end up with N separate sets of visibilities, each corresponding to a subset of the sources in the sky model. I know that this poses a problem in the source dimension - the number of sources per direction is not necessarily constant.

There are various things which could be improved upon, such as the solver being able to differentiate between quantities it must solve for per-direction vs quantities which are direction-independent and need only be computed once.

[sjperkins]

@JSKenyon This kind of thing will probably be doable in the dask version that I'm iterating on.