Draft: PostgreSQL Cluster Auto-Upgrade

Adds the ability to automatically upgrade PostgreSQL nodes in a Patroni cluster across major versions with (ideally) minimal downtime.

Upgrade step-by-step:

• be stopped
• on startup, check highest pg version in cluster
  • TODO: fix setting CONSUL_HTTP_ADDR
• if version matches our target, wait until the leader is that version
  • requires us to run a switchover/failover the first time around
• if version is above target, crash
• if version is below target, perform self-upgrade:
  • nuke data dir, initdb with target version
  • find the leader
  • TODO: find the leader's port number (patroni config?)
  • pause auto failover
  • prevent leader from stopping itself
  • for all databases on leader, dump schema and load into ours
  • temporarily switch cluster wal_level to logical
  • for all databases on leader, create replication slot for all tables, subscribe to each replication slot on ours
  • write down the current wal lsn on the leader
  • wait until we reach that lsn
  • switch the leader into read-only mode
  • sync the remaining data over, i.e. wait for replication
    • TODO: what if a ddl change happened between schema dump and read-only?
  • drop the publications and subscriptions
  • reset cluster's wal_level to previous value
  • set the cluster's sysid to our new sysid
    • TODO: is this desirable? it kills all other nodes, removing the leader
  • perform a switchover to us
    • TODO: should we do this ourselves? (currently, the second node to upgrade forces a failover on us)
  • we are now the leader
  • wait for the rest of the replicas to do the same procedure
  • cluster upgraded with minimal downtime!