datalad* orchestrators do not care to unlock outputs

[yarikoptic]

I was composing the FOI success report but decided to do the diff between sequential datalad containers-run and reproman run

using this script

```shell #!/bin/bash set -eu export PS4='> ' set -x get_participant_ids () { # Would go through provided paths and current directory to find participants.tsv # and return participant ids, comma-separated for p in "$@" .; do f="$p/participants.tsv" if [ -e "$f" ]; then awk -F'\t' '/^sub-/{print $1}' "$f" \ | sed 's/sub-//' \ | tr '\n' ',' \ | sed -e 's/,$//g' break fi done } reproman run \ --follow \ -r local \ --sub condor \ --orc datalad-no-remote \ --jp container=.datalad/containers/bids-bidsonym \ --input "sub-{p[pl]}" \ --output "sub-{p[pl]}" \ --output "sourcedata/bidsonym/sub-{p[pl]}" \ --output "sourcedata/bidsonym/dataset_description_no_deid.json" \ --output "sourcedata/bidsonym/participants_no_deid.json" \ --bp "pl=$(get_participant_ids . )" \ $PWD participant '--participant_label={p[pl]}' --brainextraction bet --bet_frac 0.5 --deid mri_deface ```

but results diverged "badly" so I realized that:

• bidsonym must not fail its run if any results saving has failed and they did fail (ATM I am rerunning one subject to verify that exit status would be 0, if it is non-0 it is somehow magically our fault in reproman although script code looks correct)
• most likely they failed because it does try to write "in-place" of the original files, which are annexed, and thus write protected.
• I have looked into reproman code and found no indication that any prepare_remote actually takes care about unlocking outputs (like it datalad containers-run does)

[yarikoptic]

per weekly chat discussion: most likely should just add a job parameter to unlock-outputs. Could possibly become not just a flag but a regex or smth like that which would "filter" corresponding to the job outputs to the ones matching or not. Behavior could be similar to datalad to simply "remove" outputs which were not provided as inputs, or we could add yet another one to be an alternative to "unlock" - remove-outputs