Improve ability to subsample/filter for creating child taxonomies

[jeremymiller]

I noticed that the subsampling/filtering of cells to create child taxonomies is not ideal. Right now subsampling for patch-seq QC is mixed up with general subsampling. For example, you can only subsample based on a column + off-target filters and a number of cells per cluster. It would be more robust if we could provide a filter/subsample vector of cells to omit or keep as an alternative to the current method. This will allow the QC filters that go into patchseqQC to be distinct from the specific cells included in a child taxonomy.