How does LTAP architecture deals with major Postgres upgrade? Is it truly zero-downtime for both upstream and downstream?