Fix fiber aperture correction: normalization and sky fibers

[rongpu]

This PR include two changes to fluxcalib:

1. Excluding sky fibers from the point_source_correction normalization calculation; sky fibers are now always assigned point_source_correction=1;
2. Changing the normalization calculation from mean to median so that it is robust against outliers.

Before this fix, fibers with very large fiber positioning offsets (mostly used as sky fibers) biases the normalization calculation, causing the FLAT_TO_PSF_FLUX value (point_source_correction in the code) to be significantly lower than 1 for the "good" fibers (see below the per-petal median FLAT_TO_PSF_FLUX value for 100 random tiles). Fortunately, downstream code cancels out this "correction" and the final flux calibration is unbiased for the good fibers. In addition to bringing the FLAT_TO_PSF_FLUX value back to a small correction around 1 as originally intended, with this PR we also fix the flux calibration of sky fibers which effectively had arbitrary normalization due to this "aperture correction".

I have tested the new code on one blue camera of a dark-time tile, and the changes to fluxcalib is very small (<1%) for all fibers except for sky fibers which now have the reasonable fluxcalib.

[coveralls]

coverage: 38.955% (+0.006%) from 38.949%
when pulling a4c3bac on rongpu-fluxcalib
into 0d7d1f2 on main.

[julienguy]

thanks!