extern thread_local is a CPU waste

In MDEV-31558 we wrongly thought that there would be minimal overhead for accessing a thread-local variable:

extern thread_local ha_handler_stats *mariadb_stats;

It turns out that in C++11, each access to an extern thread_local variable requires conditionally invoking an initialization function. In fact, we do have a custom initializer of this very variable, so that call is actually unavoidable:

thread_local ha_handler_stats mariadb_dummy_stats;

thread_local ha_handler_stats *mariadb_stats= &mariadb_dummy_stats;

It turns out that already had a better declared thread local variable that the mariadb_stats would point to when it is not pointing to the dummy object:

static thread_local THD *THR_THD;

MYSQL_THD _current_thd() { return THR_THD; }

We can access THD::handler_stats directly (well, via an accessor function, because the definition of THD is hidden from the InnoDB storage engine).