[Core] Ray auto detect nvidia Gpu with pynvml

[jonathan-anyscale]

Why are these changes needed?

We want to vendor nvidia-ml-py to Ray in order to serve several purposes:

• nvidia_gpu auto detection current implementation didn't work when Ray doesn't have root access (Ray cannot access GPUs under a non-root user (failed access of ray.init() to root-owned /proc/driver/nvidia/gpus) #28064). From several alternatives mentioned in the issue, all of them requires libnvi-ml.so which is nvidia driver C api. So, we use pynvml which is a python wrapper of this libnvi-ml.so without any other dependencies. Additionally, we can remove GPUUtil from auto detection as well.
• Removing gpustat dependencies on Ray[default] by replacing it with pynvml which gpustat depends on as well (later PR)

Related issue number

#28064

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[mattip]

This seems like the right direction, and will close #35581 and a number of issues around GPU detection.

Additionally, we can remove GPUUtil from auto detection as well.

Will this be in a follow-up PR?

[jonathan-anyscale]

This seems like the right direction, and will close #35581 and a number of issues around GPU detection.

Additionally, we can remove GPUUtil from auto detection as well.

Will this be in a follow-up PR?

The auto detection will be in this PR as well as the change is not too big + part of using nvml library. Will update once I'm done with the changes.

[jonathan-anyscale]

As per #41020 (comment)
We will move mig detection support as follow up PR

[jjyao]

@wookayin @XuehaiPan it'd be great if you can also review our usage of pynvml since you are experts here :)