[Core] Use os.sched_yield in ShmRingBuffer instead of time.sleep

[tlrmchlsmth]

While profiling #9856, I found that the calls to time.sleep in shm_broadcast.py were never taking under 50 microseconds, even if RINGBUFFER_SLEEP_INTERVAL is set to 0.

This PR switches to use os.sched_yield() instead.

Using this gist for benchmarking code, os.sched_yield() gives us close to what we wanted, at least on my system.

Benchmarking sleep(1.000000e-07) for 10000 iterations...

Results:
Requested sleep duration: 1.000000e-07s
Mean actual duration:     5.247413e-05s
Std dev:                  2.300718e-06s
Min duration:             8.199364e-06s
Max duration:             6.059092e-05s

Benchmarking os.sched_yield() for 10000 iterations...

Results:
Requested sleep duration: 1.000000e-07s
Mean actual duration:     3.195110e-07s
Std dev:                  1.182954e-07s
Min duration:             2.998859e-07s
Max duration:             1.141988e-05s

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[mgoin]

Nice find

[youkaichao]

great finding!