Fix potential race condition in lightweight_manual_reset_event.

Need to hold the mutex while calling `cv.notify_all()` to prevent
it from potentially calling `notify_all()` on a destroyed
`condition_variable` object. A thread inside `.wait()` might see
the write to `m_isSet` after the lock is released within `.set()`
but before `notify_all()` is called and then continue executing and
go on to destroy the event object (and thus the condition_variable)
leaving the thread that is about to call `cv.notify_all()` with a
dangling reference.

lib/lightweight_manual_reset_event.cpp

@@ -215,10 +215,8 @@ cppcoro::detail::lightweight_manual_reset_event::~lightweight_manual_reset_event
 
 void cppcoro::detail::lightweight_manual_reset_event::set() noexcept
 {
-	{
-		std::lock_guard<std::mutex> lock(m_mutex);
-		m_isSet = true;
-	}
+	std::lock_guard<std::mutex> lock(m_mutex);
+	m_isSet = true;
 	m_cv.notify_all();
 }