PSFs and weights come out different in g+ and g- cases

[seccolf]

When running sims with positive and negative g1=+-0.02, our expectation was that only galaxy images would change and everything else that gets saved in the MEDS files (eg weights and PSFs would be the same). That does not seem to happen.

Taking a large number of cutouts from the simulated gplus and gminus MEDS and computing ratios of the weight images then plotting the distribution of those ratios I get the plot below. The difference is fairly small (we conjecture it's simply coming from the compression of the MEDS files).

However the same exercise for the PSFs shows a larger difference, with >~1% of the pixels showing more than >~5% change in the PSFs, seen below.

[seccolf]

Things that can be checked:

1. Compression: attempt the same test above in fits files as opposed to the fpacked fits.fz
2. Unit test: output weight and psf images at the galsim stage (or really anytime before the images get written into the MEDS files). If there is already a difference in the output at that point, then obviously the MEDS stage is not the only problem.