Experiment with an inline storage for snapshot branches in postgresql

Our PostgreSQL storage currently stores snapshots in three tables:

• snapshot (id, internal id)
• snapshot_branch (internal id, name, target, target type)
• snapshot_branches (snapshot internal id, snapshot_branch internal id)

For large snapshots, this makes paginated queries (depending on the branch name) very inefficient: PostgreSQL has to fetch all branches and sort them out before outputting results.

We should experiment with a model equivalent to that of Cassandra, with one snapshot table and an inline snapshot_branches table that we can index using (snapshot id, branch name).

---

Migrated from T4671 (view on Phabricator)