fanc3_to_4 sometimes stalls? When only a single point is left?

[jasper-tms]

fanc3_to_4 usually converges in like 2 or 3 iterations (even for quite a few points) but sometimes it seems to stop making progress and then it runs all the way up to max_iterations (10) without converging. Why? Brandon thinks it's when a single point needs to be processed that the mags function gets confused. Investigate.

[jasper-tms]

After the commit above (see its commit message for more details about what was going on), fanc3_to_4 can now achieve arbitrary precision. (Well, down to 0.1 pixels of requested precision is what I tested. Smaller than that would never be useful I don't think, but I bet it will continue to work.)