Here goes the plan:
- storage1.staging: Upgrade swh.storage to v0.35 (mind the migration step)
- Make the staging check end-to-end for mercurial loader fail (as in production) by duplicating extids (rough sql magic inside).
- workers: Upgrade swh.model and swh.loader.mercurial, restart workers
- The failing staging check end-to-end should go green again
- Clean up: Drop the duplicated extids