When reaches to executor pool size, COPY sets the placement access

It looks like we forgot to set the placement accesses, and
this could lead to self-deadlocks on complex transaction blocks.

(cherry picked from commit 36bdeef1bb)

src/backend/distributed/commands/multi_copy.c

@@ -3583,6 +3583,12 @@ CopyGetPlacementConnection(HTAB *connectionStateHash, ShardPlacement *placement,
 													nodeName,
 													nodePort);
 
+		/*
+		 * Make sure that the connection management remembers that Citus
+		 * accesses this placement over the connection.
+		 */
+		AssignPlacementListToConnection(list_make1(placementAccess), connection);
+
 		return connection;
 	}