run_pcg doesn't have a convergence criteria

[skhrg]

@Sulla2012 and I noticed that right now run_pcg just runs until max_iters. It should have a stop condition when Ax - b gets to be below a threshold, the comments seem to support this.

I think we want to reduce r down to a scalar to do this but its unclear to me what the best way to do that is. Right now r is a Mapset, the most naive thing i can thing of is to dot all the maps in r with themselves and then use the sum of those values as the value to check. But we probably want something more clever (noise weighted?).

Another thing to think about is if we want the make this an absolute threshold (ie: f(r) < thresh ) or something on the change (ie: f(r_new) - f(r) < thresh). This has the advantage of not needing to worry about some offset in f(r) from having garbage at the edge of the map that isn't getting better (but if we do something noise weighted that should also take care of that I think).

@sievers do you have any thoughts on the best thing to do here?