don't zero the image at EDGE

[esheldon]

This messes up background determination, where there is some sky value everywhere but the edge

This is a problem because the sky determination code does not ignore EDGE

[esheldon]

actually the sky sub is supposed to ignore EDGE, so this might not be needed

[esheldon]

Different reason to do this: when we subtract the sky, the EDGE pixels are in fact ignored for determining the sky, but the EDGE pixels are sky subtracted by the process.

For real data we will not allow EDGE pixels to enter the coadd, but for simulated data we sometimes will for the tests without warping, so we don't want to zero the edge pixels