Fallback to restart_lsn if confirmed_flush_lsn is not found

At the moment, debezium fails against postgres 9.5 when it tries to fetch the state of the slot and it fails to find the "confirmed_flush_lsn" field in the pg_replication_slots tables which was added in PG 9.6. However, "restart_lsn" should work in most cases, put will produce extra duplicates.

In PG 9.5, we can fall back to restart_lsn and we should still be correct, but with an additional duplication of records on restart against PG 9.5