For weighted reproject_and_coadd, don't assume weight WCS is same as input

[keflavich]

Currently, the weights are treated as if they're on an identical grid to the data, but this assumption isn't enforced.

We should enforce that assumption to avoid logical errors of the variety I'm usually making - i.e., the input WCS is being applied to the weight array no matter what.

We should also make it possible to use weights that are on a different grid. This is straightforward enough, I think, but requires a good deal of checking.

[keflavich]

Fixed with 5cfa233 in #474