Handling ~1k-subject datasets (subdataset-per-subject)

# Handling ~1k-subject datasets

how to handle ~1k subject datasets like ds003097? felix made each subject a
subdataset. zipping could help too, but id like to avoid that if we can.

Note: `babs merge` already merges in chunks (default ~2000 branches), so it
may handle this scale fine as-is. TODO check that.