MDEV-16708: spurious ER_NEED_REPREPARE for gtid_slave_pos and event_scheduler

do not try to detect metadata change (and reprepare) for
internal short-lived TABLE_LIST objects that couldn't have
possibly lived long enough to see prepare and cache the metadata.

mysql-test/main/events_restart.test

 # Can't test with embedded server that doesn't support grants
 -- source include/not_embedded.inc
---disable_ps_protocol
 call mtr.add_suppression("Column count of mysql.event is wrong. Expected .*, found .*\. The table is probably corrupted");
 
 let $collation_server=`select @@collation_server`;
@@ -172,4 +171,3 @@ select name, originator, status from mysql.event;
 
 # Cleanup
 drop event ev;
---enable_ps_protocol

sql/sql_base.cc

@@ -2815,7 +2815,13 @@ static bool
 check_and_update_table_version(THD *thd,
                                TABLE_LIST *tables, TABLE_SHARE *table_share)
 {
-  if (! tables->is_table_ref_id_equal(table_share))
+  /*
+    First, verify that TABLE_LIST was indeed *created by the parser* -
+    it must be in the global TABLE_LIST list. Standalone TABLE_LIST objects
+    created with TABLE_LIST::init_one_table() have a short life time and
+    aren't linked anywhere.
+  */
+  if (tables->prev_global && !tables->is_table_ref_id_equal(table_share))
   {
     if (thd->m_reprepare_observer &&
         thd->m_reprepare_observer->report_error(thd))