Plan bidirectional metadata sync about pg_shard HOT 1 CLOSED

A quick note that could make things easier.

Citus' Select logic doesn't change a shard's (or shard placement's) metadata. So if a table is created and owned by pg_shard, we'll only need to support uni-directional metadata sync from pg_shard to CitusDB. If a table is owned by CitusDB, it's still an open question if a user would want to insert data into it through pg_shard.

Citus' Select doesn't need to change metadata because it recognizes failures in one of two ways:

1. The user updates the membership list in pg_worker_list.conf, and then sends PostgreSQL a SIGHUP. Citus won't send Select queries to nodes that aren't in the membership list.
2. If the shard placement is marked as inactive, Citus won't send a query to it.

from pg_shard.