feat: add VLLM_SPYRE_NUM_CPUS and psutil to help with cpu checks

[tjohnson31415]

Description

A couple of improvements related to setting threading based on cpu count:

• add VLLM_SPYRE_NUM_CPUS configuration to set the cpu count and skip the detection steps
  • this is useful if detection is not working but you want just 1 configuration to set and to still have vllm-spyre scale by the worker count
• adds psutil as another way to auto-detect CPUs by counting "physical" cores only instead of logical cores
  • in multi-threaded CPU bottlenecks, using logical cores may be inefficient
  • cpu detection is meant to be best-effort so I didn't add psutil as a dependency (though it does currently come in through the accelerate sub-dependency of the fp8 extras package)

Related Issues

#483

[github-actions]

👋 Hi! Thank you for contributing to vLLM support on Spyre.
Just a reminder: Make sure that your code passes all the linting checks, otherwise your PR won't be able to be merged. To do so, first install the linting requirements, then run format.sh and commit the changes. This can be done with uv directly:

uv sync --frozen --group lint --active --inexact

Or this can be done with pip:

uv pip compile --group lint > requirements-lint.txt
pip install -r requirements-lint.txt
bash format.sh

Now you are good to go 🚀

[joerunde]

Should we log here saying that psutil can be used if it's installed?

I'm also not opposed to adding it as a dependency, it seems very well maintained

[tjohnson31415]

It is well maintained, but is also OS-dependent compiled code. We could list out all of the platform_systems it does support, but 🤷. I wish there was a configuration like "try to install by default but don't stop the install if unable" 😅

I'll add a log message indicating the option to install it though!

[maxdebayser]

Yeah, perhaps that's a better option since we have to support non x86 platforms.

[maxdebayser]

bot:test

[tjohnson31415]

The spyre-ci failure was unrelated to my changes. It was fixed in #491, so I rebased this PR on main.

[tjohnson31415]

bot:test