Improvement: stop MQTP queues in parallel

[678098]

bmqc::MultiQueueThreadPool::stop iterates over all processing queues and enqueues a monitor event to request stop, then wait each queue to stop serially.
It might be more efficient to split this loop in two loops.
In the first loop, send a monitor event to the queue and disable push.
In the second loop, wait for all queues to finish using the shared TimedSemaphore.

Implementing this causes some tests to fail. We might have hidden assumptions how the stop is done, and it's worth investigating.

The code:

blazingmq/src/groups/bmq/bmqc/bmqc_multiqueuethreadpool.h

Lines 1161 to 1182 in 47a89d5

	for (size_t i = 0; i < d_queues.size(); ++i) {
	QueueInfo& info = d_queues[i];
	QueueItem item = {false // isMonitorEvent
	,
	0}; // pointer to event
	info.d_queue_p->pushBack(item);
	info.d_queue_p->disablePushBack();
	// TODO: Change 'd_processedZero' to semaphore instead to avoid busy
	// waiting?
	while (!info.d_processedZero) {
	bslmt::ThreadUtil::yield();
	}
	while (!info.d_queue_p->tryPopFront(&item)) {
	d_pool.releaseObject(item.d_event_p);
	}
	d_allocator_p->deleteObject(info.d_queue_p);
	info.reset();
	}