When reaches to shared 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.

src/backend/distributed/commands/multi_copy.c

@@ -3649,6 +3649,12 @@ CopyGetPlacementConnection(HTAB *connectionStateHash, ShardPlacement *placement,
 			connection =
 				GetLeastUtilisedCopyConnection(copyConnectionStateList, nodeName,
 											   nodePort);
+
+			/*
+			 * Make sure that the connection management remembers that Citus
+			 * accesses this placement over the connection.
+			 */
+			AssignPlacementListToConnection(list_make1(placementAccess), connection);
 		}
 		else
 		{