Reduce inode usage in `participant_job.sh`

[tien-tong]

Summary

In participant_job.sh, the first step is a full-tree datalad get -n of the input dataset and then prunes other subjects/sessions with find/rm -rf:

# pull down input data (but don't retrieve the data content) and remove other sub's data:
echo "# Pull down the input subject (or dataset) but don't retrieve data contents:"
datalad get -n "inputs/data/BIDS_dl/${subid}"
(cd inputs/data/BIDS_dl && find . -type d -name 'sub*' | grep -v "$subid" | xargs rm -rf)
(cd inputs/data/BIDS_dl/"${subid}" && find . -type d -name 'ses*' | grep -v "$sesid" | xargs rm -rf)

Even without retrieving file contents, this still materializes the full directory tree for the whole input RIA (all subjects/sessions). On large BIDS datasets this consumes too many inodes, and we repeatedly hit inode limits on shared filesystems.

Additional details

Next steps