Add hidden replica identity to shards

When a distributed table does not have a replica identity, UPDATE and DELETE commands error out if we use logical replication. This prohibits us from using logical replication in a structural way (e.g. background rebalancing), and worsens the user experience for one-offs such as rebalancing using logical replication.

To not force the user to create a replica identity, we could add one automatically by adding a hidden column (#1728) which is populated using a sequence.

The downside of the hidden column is a performance and storage-size penalty, especially from the unique index. We should allow the user to disable hidden replica identity on a table.