Update centroid while iterating FWHM calculation

Comment from Robert on Slack (auto-guiding channel)

While you're looking at the windowedFWHM code, I think it'd be better to update the centroid within the iteration.  You can use the Gaussian weighted moment of x and y to do this;  I think it'll converge faster if you adjust the centroid by twice the error, but with your 30 iterations this shouldn't matter (you'll have to add the centroids to the convergence criteria of course).

This wouldn't be necessary if the image quality were better for the out-of-focus images at the edge of the focal plane, but some of them really are pretty nasty and this would be safer.