Normalise and de-duplicate source and target branches for merge requests

In #331523 (comment 583693337) we found we can save quite a bit of space by not storing source and target branches as separate values for every occurrence. Instead, we'd store the unique values in a separate table much like discussed in #331823 (closed).

The logic for migrating and populating data at runtime would be more or less the same (if not the same), so please take a look at that issue for more information.