Once we've detected Sources in multiple bands we need to merge the positions to generate Objects.

This is a little complicated (or at least messy):

• The positions have errors
• If the seeing is different in different visits, objects may be blended in some but not all exposures
• If we use more than one detection pass (e.g. smoothing when looking for faint objects, not smoothing for bright) this has similar-but-different consequences (but we should probably deal with in the per-band processing)
• Objects move, so even if the positions are within the errors the motion may still be detectable