[BUG] Fscale in ngmix_runner

[aguinot]

To handle the different zero_point on single exposures while doing the multi-epoch processing, the images need to be rescale. The same rescaling needs to be apply on the weight to be consistent. This done at this line. But latter the weights are to 1 before being recomputed and then we lost the rescaling.

We should remove the part where the weights are set to 1. Also, the weights should be computed outside the ngmix_runner. See #604

Thanks to @lbaumo for finding that!

[aguinot]

Ok, false alarm 😮‍💨
So it is very badly handle in the code but there is no problem actually. The noise variance is measured on the galaxy stamps that are scaled so it is propagated to the weights. It is really hidden in the code and something like that should appear very clearly! (to avoid exactly this kind of issues)

To fix this, as mentioned in issue #604, the weights should be measure outside ngmix. So we can keep the line 432 and remove the line 816 where the weights are set to 1 (of course the line 860 will also need to be removed.)

I am closing this issue since there is no actual "bug" in the code as it is but a poorly implementation 😕