automatically keep certain covariates in making the pseudobulked cds

[maddyduran]

No description provided.

[rfriedman22]

Some of the CCS colData transfers over incorrectly since the colData is on the embryo level and the left join is done by sample. For example, in a (very old) muscle CDS I am working with, I know that cluster 8 is slow muscle and cluster 9 is fast muscle, but they are both given cell_type_broad of "mature slow muscle" and cell_type_sub of "mature fast muscle 1."

The fact that two clusters with different cell type broad labels are given the same label, and that the cell type broad label is nonsensical compared to the cell type sub label, indicates there is something strange going on with how the join is done. I haven't investigated further yet.

There need to be some policies to better transfer this metadata over. This isn't mission critical for anything I'm doing but it can cause users to misinterpret their pseudobulk.