bug in decreasing allowableThreads

Nobody has claimed this yet.

Decreasing allowableThreads seems to take a full cycle before taking effect. I don't think this behavior is documented, and could potentially be detrimental:
```m2
i1 : allowableThreads = 8;

i2 : elapsedTime scan(apply(10, i -> schedule(() -> sleep 1)), taskResult)
-- 2.00075s elapsed

i3 : allowableThreads = 4;

i4 : elapsedTime scan(apply(10, i -> schedule(() -> sleep 1)), taskResult)
-- 2.00046s elapsed

i5 : elapsedTime scan(apply(10, i -> schedule(() -> sleep 1)), taskResult)
-- 4.00064s elapsed
```
Notice that the second time around, the correct number of threads were used, hence the total time was doubled. This doesn't seem to happen when increasing allowableThreads.

Reproduce the timing sequence using allowableThreads, schedule, sleep, and taskResult, first confirming the delayed effect when decreasing the limit. Then locate the implementation of allowableThreads and trace how changes are applied across scheduled work. Done means the observed behavior is corrected or its intended semantics are made explicit, with the reproduction covered by a suitable regression check.

Written by the indexing model from the issue text.