[profiler] fix multi-aiu profiling and add setable options

[mcalman]

Description

1. Switches to an alternative method for enabling AIU event profiling when profiling is enabled with VLLM_TORCH_PROFILER_DIR. This resolves the following error that occurs when trying to profile a multi-AIU workload: RuntimeError: Please register PrivateUse1HooksInterface by 'RegisterPrivateUse1HooksInterface' first.
2. Adds support for the following profiling options that were added to vLLM: https://github.com/vllm-project/vllm/blob/30f78af147cb9eac0a5c643a69882a3b0e45f986/docs/contributing/profiling.md?plain=1#L10-L13

Related Issues

n/a

[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 🚀

[maxdebayser]

What is the use case for setting VLLM_SPYRE_AUTOPILOT_OFF instead of removing autopilot from DT_OPT?

[mcalman]

This is provided for convenience and to make user's aware the option to turn autopilot off exists and is often needed for informative profiling. It is functionally equivalent to the user setting DT_OPT with the desired autopilot value themselves.

[maxdebayser]

I see. I agree that the DT_OPT environment variable is not very convenient, but I think that creating these interactions between variables could make things more confusing for the user. What if instead whenever envs.VLLM_TORCH_PROFILER_DIR is set and the autopilot option in DT_OPT is not set a warning is printend? In this way the user would be made aware as well.

[mcalman]

@maxdebayser thanks for the suggestion. I've made this change