BUG#34045080 Failure with the THD::mem_cnt pointer

Post merge fix, to relax an assert.

Thd_mem_cnt::enable() no longer require that THD
is associated with some performance schema instrumentation,
because this can in fact not be enforced:
the THD <--> performance_schema thread instrumentation
can change over time, so testing that a performance schema
thread instrumentation actually exists when Thd_mem_cnt::enable()
is called goes not give any guarantees about the existence
of a thread instrumentation when Thd_mem_cnt::alloc_cnt()
is called.

This fix relaxes the asserts in Thd_mem_cnt::alloc_cnt(),
because there is no guarantee that a performance schema
thread instrumentation is in place for the THD used.

This is especially important for replication,
when a replication thread A can pick up a different session B
and operate on B on its behalf, see Thd_backup_and_restore()

Approved by: Tor Didriksen <tor.didriksen@oracle.com>

Author: marcalff

sql/sql_class.cc

@@ -166,7 +166,7 @@ bool Thd_mem_cnt::alloc_cnt(size_t size) {
     return false;
   }
 
-  assert(!opt_initialize && m_thd->get_psi() != nullptr);
+  assert(!opt_initialize && m_thd != nullptr);
   assert(!m_thd->kill_immunizer || !m_thd->kill_immunizer->is_active() ||
          !is_error_mode());
   assert(m_thd->is_killable);