ntdll: make NtDelayExecution a bit more efficient

[Graham]

Alexandre wrote:
> That's what the existing code already does.

Indeed. I don't know what I was thinking...

>> 2. If you're about to block on select(), then I don't see any point in
>> preceding that with a call to sched_yield().
>
> This was added for a reason; most likely you'll have to write tests.

i.e. commit 8099c2b9. JW says "... to more closely resemble Windows
behavior.  The key is to yield in a Sleep..."

I haven't yet figured out how he came up with this analysis, but I
think it's safe to assume that he is correct, and my patch is garbage.
Lesson learned: consult history.

-- graham