To load diffs and other files, we make two SELECT calls in
`MergeRequest#environments_for`: one for the source branch, and one for
the target branch.

The source branch query makes a subquery:

```sql
SELECT environment_id FROM deployments WHERE project_id = 123 AND ref =
'source_branch'
```

The target branch makes this subquery:

```sql
SELECT environment_id FROM deployments WHERE project_id = 123 AND (ref =
'master' OR tag IS TRUE)
```

Both subqueries were taking over 250 ms to execute. When we add an index
for the deployments table on `(projects_id, ref)`, this execution drops
down to under 5 ms since the table usually has a small number of rows
for the source branch.

Adding a partial index on `projects_id WHERE tag IS TRUE` allows
PostgreSQL to perform two separate indexed scans: one for the ref, and
one for the tag condition. However, this query still takes 60 ms to run
for the gitlab-org/gitlab project because `master` has 10,000+ rows. We
should consider adding a LIMIT in a future iteration.

Closes https://gitlab.com/gitlab-org/gitlab/issues/55353