Since this worker can update the same row for a group or project,
retrying can exacerbate locking in the database. PostgreSQL may create a
MultiXactId as a side effect with each lock. If these IDs exceed the
working memory of the simple LRU (SLRU) cache, database queries could
get slow.

If we are unable to store the data after 3 tries then we're better off
giving up and thereby reducing contention.

Relates to https://gitlab.com/gitlab-org/gitlab/-/issues/340272

Changelog: performance