Synthetic.jl image generation, asinh magnitudes, and various fixes

[gostevehoward]

Sorry this PR got so large, but here are the highlights:

• Implements generation of FITS imagery from a catalog CSV using Synthetic.jl (fixes #619)
• Replaces Pogson magnitudes with asinh magnitudes (fixes #618)
• In both GalSim and Synthetic.jl FITS generation, use a WCS which matches that used in SDSS imagery as closely as possible (see https://github.com/jeff-regier/Celeste.jl/wiki/Coordinate-systems for the gory details)
• Modify "epsilon" and "iota" in both GalSim and Synthetic.jl image generation to more closely match SDSS imagery (but not in GalSim benchmark image generation)
• When matching SDSS primary predictions to SDSS coadd predictions, limit position matching radius to 1 px (rather than 1 / .396 ~ 2.5 px). This was the original intention, it doesn't cut down on number of matched sources much, and it will hopefully avoid some bad matches which throw misleading outliers into the error analysis.
• Add various features to the benchmark/accuracy suite to support running on other SDSS RCFs and to aid debugging (e.g., writing raw matched predictions and raw normalized errors to CSVs)

[gostevehoward]

Found and fixed the bug with angles in Synthetic.jl imagery (just a radians/degrees bug).

I found it by making this catalog (by hand, in Google Sheets) and generating FITS imagery from it, which looks like this (after the fix):

The benchmark/accuracy suite makes it easy to set up and check stuff like this.

[jeff-regier]

Looks great Steve. Thanks for your work here.