Revisit pathman readiness again.

Most pathman functions break if pathman is disabled: earlier I put defense
checks in split and merge range partitions, now it is
drop_range_partition_expand_next. Looks like the reason is that pathman caches
are not being invalidated if it is disabled: pathman_relcache_hook exits right
away then. This is kinda reasonable: if we want to disable pathman completely,
why maintain the caches? So this time try to bury the readiness check deeper,
in get_pathman_relation_info itself.

BTW, pathman caches are not dropped when it is disabled, which looks suspicious
on its own -- probably if we re-enable it later, caches might be inconsistent.

Author: arssher

expected/pathman_calamity.out

@@ -1009,7 +1009,7 @@ ERROR:  pg_pathman is disabled
 SELECT * FROM pathman_partition_list;							/* not ok */
 ERROR:  pg_pathman is not initialized yet
 SELECT get_part_range('calamity.survivor', 0, NULL::INT);		/* not ok */
-ERROR:  pg_pathman is not initialized yet
+ERROR:  pg_pathman is disabled
 EXPLAIN (COSTS OFF) SELECT * FROM calamity.survivor;			/* OK */
           QUERY PLAN          
 ------------------------------

src/pl_range_funcs.c

@@ -489,9 +489,6 @@ split_range_partition(PG_FUNCTION_ARGS)
 	char			   *query;
 	int					i;
 
-	if (!IsPathmanReady())
-		elog(ERROR, "pg_pathman is disabled");
-
 	if (!PG_ARGISNULL(0))
 	{
 		partition1 = PG_GETARG_OID(0);
@@ -655,9 +652,6 @@ merge_range_partitions(PG_FUNCTION_ARGS)
 	FmgrInfo			finfo;
 	int					i;
 
-	if (!IsPathmanReady())
-		elog(ERROR, "pg_pathman is disabled");
-
 	/* Validate array type */
 	Assert(ARR_ELEMTYPE(arr) == REGCLASSOID);
 

src/relation_info.c

@@ -320,6 +320,9 @@ get_pathman_relation_info(Oid relid)
 {
 	PartStatusInfo *psin;
 
+	if (!IsPathmanReady())
+		elog(ERROR, "pg_pathman is disabled");
+
 	/* Should always be called in transaction */
 	Assert(IsTransactionState());