Attempt to fix thread_pool_scheduler test

[msimberg]

Added predicates to the wait_fors. Without them the waits may fail even if the the tasks were executed (and since

hpx/libs/parallelism/executors/tests/unit/thread_pool_scheduler.cpp

Line 1229 in 0bfa445

HPX_TEST(called);

seems to not fail, it must be the case here).

[hkaiser]

LGTM, even if I was not aware that we have spurious wakeups.... Thanks, I would have never thought of applying this change!

[msimberg]

LGTM, even if I was not aware that we have spurious wakeups.... Thanks, I would have never thought of applying this change!

The failure doesn't require spurious wakeups (and I don't think they happen in the runtime). It just requires that the task notifies before the wait happens (either the task happens to be scheduled immediately, or more likely that the waiting thread happens to give up its timeslice just before it waits).

Note: I wasn't able to reproduce this locally, so I'm only guessing that it fixes the problem. The symptom and fix does fit CircleCI being heavily oversubscribed though (and the fix is needed in any case).