Introduce citus_pause_node

Add a UDF that takes the same locks as citus_update_node, but doesn't do anything else. This is useful for a coordinated restart of a worker node. If a connection calls citus_pause_node any queries to that node would be held until the connection is released. This reduces the number of errors during a restart, by changing them from errors to higher latency.

1. BEGIN;
2. citus_pause_node(A)
3. restart node A
4. COMMIT/ROLLBACK