The background migration MigrateMergeRequestDiffCommitUsers was wrapping
its updates in a transaction. This was a left-over from an early draft
of the merge request. As the bulk updates can take some time to
complete, this can lead to long transaction timings. On production this
ultimately lead to the incident described in issue
https://gitlab.com/gitlab-com/gl-infra/production/-/issues/5119, though
fortunately this didn't seem to negatively affect users.

This commit removes the transaction, and adds a migration to reschedule
all the background jobs (which have been removed from the queue).
Rescheduling all jobs is the simplest solution to ensuring all data is
migrated. Rows that have already been migrated will be updated again,
but that's not a problem. Any existing data in
merge_request_diff_commit_users will just be reused.

See the following for more details:

- https://gitlab.com/gitlab-org/gitlab/-/merge_requests/63669#note_621297454
- https://gitlab.com/gitlab-com/gl-infra/production/-/issues/5119
- https://gitlab.com/gitlab-org/gitlab/-/issues/331823

Changelog: fixed