Roman Mukhin

Perhaps our more general solution that based on `anndata._io.specs import read_elem` could address this use case as well https://pypi.org/project/cap-anndata/

I've found that RAM usage depends on concurrency settings:  Maximum RAM usage stops growing after 4, because there are 4 shards to read. So far, I assume that each...

I've found that setting concurrency to 1 (`os.environ["ZARR_ASYNC__CONCURRENCY"] = "1"`) solves the issue.