I believe most major versions of postgres only change metadata, so the downtime only scales with the size of the schema (usually small) and not the size of the data.

You can do live parallel routing and blue-green migrations transparently. In your docs you already describe how to do it but the user has to do it manually at the application level when you can do it just fine at the DB load balancer/router level.