Decouple distributed table schema from shard schema

We currently assume that the table schema of shards matches that of the distributed table exactly. In the future, we may want the ability to add hidden columns to shards in order to track certain information such as the distributed transaction ID of a row, or an implicit primary key (for logical replication)

Because existing distributed tables would not have hidden columns in the shards, we may want to implement a versioning scheme for distributed tables (#1727).