gh-110733: Optimize _run_once for many iterations of the event loop

[bdraco]

Replace min and max in _run_once with simple < and >

close #110733

• Issue: Reduce overhead to run one iteration of the asyncio event loop #110733

benchmarks in #110733

new: 0.36033158400096
original: 0.4667800000170246

[gvanrossum]

The patch LGTM. I haven't confirmed the speedup but I agree that not calling min() or max() is slightly faster. It will matter even more in 3.13 when we expect to have a basic tracing JIT.

I noticed that you considered calling time() only once -- do you want to add that to the PR? Or did you already measure it and it doesn't help?

[bdraco]

I noticed that you considered calling time() only once -- do you want to add that to the PR? Or did you already measure it and it doesn't help?

I couldn't come up with a way to get it to work that wasn't a breaking change since we need the time before the select and the time after the select in most cases because time moves forward during the select

[bdraco]

Thanks!