[V1] Simplify Shutdown

[robertgshaw2-neuralmagic]

SUMMARY:

• objects that for manage background processes or IPC implements shutdown and weakref.finalize
• remove use of __del__ (resolves weird logs when LLM is cleaned up
• remove need for higher level classes like LLM, LLMEngine, or AsyncLLM to need to do anything special for shutdown, since all the objects responsible for managing their IPC and MP resources handle their own shutdown

[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.

🚀

[robertgshaw2-neuralmagic]

@WoosukKwon - this solves the logs you were seeing

[tlrmchlsmth]

LGTM

Note: I'm seeing some shared memory leaks when running examples/offline_inference.py with tensor_parallel_size=2 that we should fixup:

/usr/lib/python3.10/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 1 leaked shared_memory objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

(Not seeing this when using vllm serve or when running with tensor_parallel_size 1)