Add superslab filtering functionality

[lgarrison]

This PR introduces a feature for filtering rows from the halo table on a per-superslab basis as they are loaded into memory. This saves memory, since the full, unfiltered table never needs to be constructed. Syntax is a function passed to the CHC constructor, often a lambda:

cat = CompaSOHaloCatalog('/mnt/home/lgarrison/ceph/AbacusSummit/AbacusSummit_hugebase_c000_ph000/halos/z0.100/',
                         fields=['N','x_L2com'],
                         filter_func=lambda c: c['N'] >= 100,
                        )

If requesting subsamples, only the subsamples from the surviving halos are loaded, but the peak memory usage is still higher than it needs to be. This part of the code needs a refactor anyways (#7), but that's a future issue.

This required a bit of refactoring of how the cleaning files are read; now the original and cleaned files are read as a pair for each superslab, rather than all the originals, then all the cleaning.