min_n_subfiles could be better

[christinawlindberg]

For ACCESS BEAST runs, subdividing files based on min_n_subfiles can be problematic because, if min_n_subfiles > 1, then if a catalog bin has very few sources, there's a decent chance that one of the subfiles will have just a single source, generating an error when fitting i.e. #762.

On the other hand, if min_n_subfiles = 1, then bins with few stars (<1000 based on n_per_file) won't be split into subfiles, meaning that the observational physics grid size remains huge.

Potential Solution:

1. Implement a max_subfile_size limit in the BEAST settings, where, after splitting files, the code checks to see if any of the noisemodels or trimmed sed grids exceed the max_subfile_size limit.
2. Subfiles are currently sorted and split according to "F475W". In order to optimize the splitting of the catalog, it might be beneficial to re-sort the subfile catalogs by a second filter e.g. F336W or F814W, before splitting a second time.