model CCD saturation as censoring

[jeff-regier]

For bright stars and galaxies, the CCD can't record all the photons it receives. Pixels "fill up" around the brightest parts of the light source. Whether a light source saturates has to do with its brightness, of course, but also the effects of "seeing." (With a wider point spread function, photons are spread across more pixels, perhaps preventing saturation.) It's very easy to detect saturation because a lot of contiguous pixels will all contain nearly some maximum value. By modeling the observed pixel value as censored, we should be able to accurately recover bright light sources' magnitudes.