Fix TwoDHist plot and GSObject fitting

[rmjarvis]

This PR has two extremely small fixes that don't have anything to do with each other, besides that they were prompted by runs on Roman coadd images by Brenna (@WellsBN).

1. We noticed when looking at the TwoDHist outputs that the T_model panel was actually just a copy of the T panel. This was a straightforward copy/paste bug in the code, which is fixed here.
2. When fitting the Gaussian profile, she was sometimes getting crashes that seemed to be out-of-memory errors, from extremely large memory allocations. I'm pretty sure this is because scipy was sometimes trying bad parameter combinations in the non-linear fitting. This used to work fine, since GalSim would throw an exception when the FFT was too large, and Piff would catch those errors and call the proposed parameter bad. But now the default behavior is to just warn for that rather than error. The fix is to just change GalSim's default back to errors, which is a simple global variable edit.

[nihardalal]

looks good to me!