LWG-3343 Ordering of calls to `unlock()` and `notify_all()` in Effects element of `notify_all_at_thread_exit()` should be reversed

[StephanTLavavej]

LWG-3343 Ordering of calls to unlock() and notify_all() in Effects element of notify_all_at_thread_exit() should be reversed

From a quick glance, I believe it's these lines, but I haven't investigated thoroughly:

STL/stl/src/xnotify.cpp

Lines 96 to 97 in ab6dec6

	_Mtx_unlock(block->data[i].mtx);
	_Cnd_broadcast(block->data[i].cnd);

[vmichal]

For what it's worth, I can confirm that swapping those two lines should fulfill the LWG issue.
However, the other part of the linked issue

This notification is sequenced after all objects of thread storage duration associated with the current thread have been destroyed

makes my head spin a little. The lifetime of objects in TLS is not immediately apparent to me, so I would appreciate suggestions on how to verify that this requirement is fulfilled as well.