Sersic sources

[JelleAalbers]

This allows Manada to generate Sersic sources similar to those in the COSMOS catalog. It is intended to help diagnose potential problems in the inference stage, and perhaps facilitate comparisons to earlier papers.

In more detail, this:

• Adds COSMOSSersicCatalog, which behaves like COSMOSCatalog except that, rather than actual observed images, it produces best-fit single elliptical Sersic profiles (according to the 'SERSICFIC' field in the COSMOS metadata).
• Adds a configuration config_d_los_sigma_sub_smooth.py which imports all settings from config_d_los_sigma_sub, but switches the catalog class to COSMOSSersicCatalog.
• Adds a configuration config_supereasy.py which does whatever it can to make the sigma_sub inference as easy as possible. It generates no noise, assumes the PSF is a delta function, halves the pixel length (quadruples the image area), and turns off all line of sight substructure.
• Moves the manada Configs directory inside the manada package. Currently they are in a separate directory outside the package, which means we can't (easily/nicely) import one config file from another.
• Add a flag --save_png_too to generate.py, which causes it to save a PNG image along with every npy file. It is not a new config option since it doesn't change what images/metadata are produced, it's just for quickly checking what a config is doing.

Here are some example images to compare the different smoothing procedures. The first column is the full original image, all other columns are 64x64 images from manada (config_d_los_sigma_sub), produced with a zero-arcsec theta_E lens, no random rotation, and a fixed seed for noise generation. The second column uses smoothing_sigma = 0, the third the default (smoothing_sigma = 0.08, and the final column uses COSMOSSersicCatalog as the source class (for which smoothing_sigma has no effect).

(@swagnercarena I see the PR picked up your unmerged commits from parameter_changes -- let me know if you'd rather merge those separately, or if you'd like me to rebase this PR on master)

[swagnercarena]

This all looks great! I pulled the changes from parameter_changes into master, so that should no longer be an issue. I just added two tests for the individual functions (a little picky on my end but I like to have each function have its own test when possible). I also added a bit more documentation. Feel free to merge with master if everything looks good to you.

[JelleAalbers]

Thanks Sebastian, the extra tests look good! You even added docstrings for functions that just raise NotImplementedError :-)