Speed up saturated registration mode

[arjunsavel]

Is your feature request related to a problem? Please describe.
The saturated image registration mode is currently orders of magnitude slower than the default mode. If it were faster, though, its level of robustness would make it the clearly preferred registration method.

Describe the solution you'd like
Using Cython or Numba would probably be best.

Describe alternatives you've considered
Parallelizing the code might be helpful, but not sure if this is too dependent on the number of cores available to the user.