Apparently, some 0.2% of our DispatchChanges jobs are submitted but just never run (see T292609). While the exact cause is still unclear, we need some way to deal with it.
We need to achieve the following:
- the subscribed page at the client wikis should be purged at some point
- the change should be injected in the recent changes and watch lists at the client wikis
- the row should be removed at wb_changes
- things should not explode in case the jobs lag a bit
- the number of job rescheduled that way should be tracked
Acceptance criteria:
- write a maintenance script that resubmits jobs for entries in wb_changes that are older than 24 hours
- run this maintenance script with a cron job regularly (specific interval TBD)
