Replace short-period coadds with coadd difference images

Instead of producing short-period (e.g. yearly) coadds for the purpose of detecting and characterizing faint, slowly-changing sources, I believe we should diff these coadds against a full-depth coadd (or perhaps each other), and detect and characterize on the difference images.

This should reduce blending issues in characterizing these objects, and it should make what I'm calling the DeepAssociate pipeline (where we merge Footprints and Peaks to define Object candidates, just before deblending) much easier, because the short-period coadds will no longer contribute redundant detections for static sources at single-year depth.

I thus far have not included this proposal in LDM-151, in order to maintain consistency with the DPDD.