Amends commit 5b5297fe87859f59a7aaf5e86a8915c00714fefa to make it more reliable against CI timeouts and flakiness. Now, we wait for the threads that are expected to timeout to actually timeout before releasing the target thread they were waiting on. It should be impossible now for this to go wrong. There should also be no problem of their handing off to a thread that will wait forever. I've added two more tests that could, possibly, have a problem in the CI: when timing-out threads hand off to a timed wait that is expected to succeed (the > 1s wait cases). Though this should be a rare occurrence, if ever: the target thread's runtime is the longest of the timing out threads' wait, plus 5 ms. That is 30 ms in the examples I wrote, so the additional extra time they're waiting on should be more than enough. An extra benefit is that this test now runs much faster, at 10 to 60 ms per test row, instead of 800 ms previously. The drawback is that a failure condition is likely going to be noticed by the QSemaphores deadlocking. Change-Id: I360d7e9c7accc1216291fffd743c88a362cf66ac Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>