Allow option for users to run a numba-based interpolator that scales with number of cores

[tyler-a-cox]

No description provided.

[steven-murray]

@tyler-a-cox this is great!

Here's some thoughts:

1. We should probably write our own fast_interp unless we want to support that package ourselves into the future (which sounds difficult)
2. The interface to matvis in the tests obviously needs to be updated
3. I do wonder if we could do better by putting a simple multiprocess map (or mpi map or whatever) outside the time-loop. Everything within is embarrasingly parallel, so I would guess that the overhead would not be significant, especially for a long-time-axis simulation. We could even try to do the best of both worlds, so have an outer map and let each subprocess have two cores, which maximises the core-to-speedup ratio for each subtask.

[tyler-a-cox]

Closing this in favor of #29