Improve bad facet identification

[twshimwell]

We attempt to recognise bad facets by the errors in the astrometry derived from PanStarrs. Presently we use a criteria of 0.5" and if the error exceeds this then during mosaicing (in mosaic.py) we blank the facet. This approach needs revising because its been found that this criteria sometimes removes perfectly good facets particularly at low declination or low galactic lattitude.

In offsets_compact branch the new script find-bad-facets.py attempts to detect poor facets using the image_full_low_m image by looking at a number of different things including: properties of a Gaussian fitted to the facet pixels, trying to identify jumps in the noise properties at facet edges, the dynamic range and the astrometry. Each facet then gets a score for how bad it is depending on these properties and hopefully poor facets can be moderately robustly identified.