Fix bug in fiber aperture correction that affects cframe IVAR

[rongpu]

The calibration vector ccalibration is normalized by the fiber aperture correction factor FLAT_TO_PSF_FLUX (point_source_correction in the code), but its inverse variance was not normalized accordingly. This PR fixes that.

I tested the new code on a B camera for a dark tile. For most fibers with good fiber positioning (FIBERSTATUS==0), the effect should be negligible (less than 1% change in the average cframe IVAR). However, for very bright sources (standard stars and particularly bright galaxies), the new IVAR can be on average ~40% lower.

The largest effect is on the stuck positioners used as sky fibers. They have particularly large FLAT_TO_PSF_FLUX due to their large DELTA_X/Y, and combined with this bug their IVARs are much underestimated. For example, below is the signal-to-noise (flux * np.sqrt(ivar)) of a sky spectrum from such a stuck positioner (fiber 179 of exposure 235175; it is 140 micron "off target" and it has FLAT_TO_PSF_FLUX of ~28 compared to the typical value of ~0.7). As expected, the distribution becomes a standard Gaussian with the fix.

[coveralls]

coverage: 38.995% (+0.001%) from 38.994%
when pulling 7102957 on ccalibivar-fix
into e033a8c on main.

[julienguy]

Good catch. I note that this correction affects the uncertainty on the calibration vectors that is a small contribution to the total uncertainty for most targets. It's only affecting the brightest sources where the calibration error starts to matter, and peculiar fibers where we make a huge fiber aperture loss correction.