fix: contextvars propagation in ThreadPoolExecutor calls

[galkleinman]

When using ThreadPoolExecutor to run LLM calls concurrently with
guardrails, contextvars were not being propagated to the worker thread.
This caused issues with trace context and other context-dependent
functionality being lost across thread boundaries.

The issue broke the traces of: https://github.com/traceloop/openllmetry

Changes:

• Use copy_context() before submitting tasks to ThreadPoolExecutor
• Apply fix to chat.completions.create, responses.create, and responses.parse
• Add tests to verify contextvars propagate correctly across threads
• Add tests for single and multiple contextvars propagation

[gabor-openai]

@codex pls review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[gabor-openai]

Thanks so much for this @galkleinman ! Clean, succinct, and includes relevant testing. @steven10a to take a final look before we merge.

[steven10a]

LGTM! Thank you for catching this and providing the fix

[galkleinman]

Thanks for the quick turnaround @gabor-openai, @steven10a 🙌