fix: provider_data_var context leak

[jaideepr97]

What does this PR do?

Following PR description is generated using claude:

PR #5168 fixed OTel trace context leaking into background workers, but PROVIDER_DATA_VAR — the ContextVar that carries authenticated user identity — suffers from the same asyncio.create_task copy semantics. When a background worker is spawned, it permanently inherits the spawning request's PROVIDER_DATA_VAR, causing all subsequent DB writes to be stamped with the wrong user's identity. In multi-tenant deployments with auth enabled, this means:

• Chat completions written through the InferenceStore write queue get attributed to whichever user's request first triggered worker creation, breaking row-level access control via AuthorizedSqlStore.
• Responses processed through the OpenAIResponsesImpl background worker pool run under the wrong user's identity, affecting status updates, error handling, and stored response ownership.

This PR generalizes the OTel-only utilities from #5168 into a unified RequestContext that captures both the OTel trace context and PROVIDER_DATA_VAR together. The three helpers in core/task.py are replaced:

Before (#5168)	After (this PR)
capture_otel_context()	capture_request_context() — snapshots OTel context and provider data
activate_otel_context(ctx)	activate_request_context(ctx) — restores both per work-item
create_task_with_detached_otel_context(coro)	create_detached_background_task(coro) — clears both before task creation

Both InferenceStore and OpenAIResponsesImpl are updated to capture a RequestContext at enqueue time and activate it in the worker loop, ensuring each work-item runs under the correct user identity and trace.

Closes #5221

Test Plan

• tests/unit/core/test_task.py (10 tests): Verifies RequestContext capture/activate semantics, detached task isolation for both OTel and PROVIDER_DATA_VAR, caller context restoration, queue-based propagation patterns, and cross-contamination prevention.
• tests/unit/utils/inference/test_provider_data_leak.py (1 test): Reproduces the InferenceStore write queue leak end-to-end — two users store completions through the async queue, then verifies each user can only see their own completions via AuthorizedSqlStore access policies. This test fails without the fix.
• tests/unit/providers/agents/builtin/test_responses_background.py (6 new tests):
  • TestResponsesOtelContextPropagation (3 tests): Verifies OTel trace attribution through the responses background worker — each response is processed under its originating request's trace, contexts don't leak between items, and error handlers run under the correct trace.
  • TestResponsesProviderDataPropagation (3 tests): Verifies user identity propagation — each response runs as the correct user, identity doesn't leak between queue items, and error-handling DB writes use the correct user.

[mergify]

This pull request has merge conflicts that must be resolved before it can be merged. @jaideepr97 please rebase it. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[iamemilio]

LGTM. This was a good catch. Thanks!

[cdoern]

hmmmm looking at this, I wonder if this would've been useful to be in the API pkg if used by providers... not something to change in this PR though.

[jaideepr97]

@Mergifyio backport release-0.6.x

[mergify]

backport release-0.6.x

☑️ Command disallowed due to command restrictions in the Mergify configuration.

Details

• sender-permission >= write

[jaideepr97]

@cdoern please backport this

[leseb]

@Mergifyio backport release-0.6.x

[mergify]

backport release-0.6.x

✅ Backports have been created

Details

• #5247 fix: provider_data_var context leak (backport #5227) has been created for branch release-0.6.x but encountered conflicts

Cherry-pick of 9b86ce8 has failed:

On branch mergify/bp/release-0.6.x/pr-5227
Your branch is up to date with 'origin/release-0.6.x'.

You are currently cherry-picking commit 9b86ce80.
  (fix conflicts and run "git cherry-pick --continue")
  (use "git cherry-pick --skip" to skip this patch)
  (use "git cherry-pick --abort" to cancel the cherry-pick operation)

Changes to be committed:
	new file:   tests/unit/utils/inference/test_provider_data_leak.py

Unmerged paths:
  (use "git add/rm <file>..." as appropriate to mark resolution)
	deleted by us:   src/llama_stack/core/task.py
	both modified:   src/llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py
	both modified:   src/llama_stack/providers/utils/inference/inference_store.py
	deleted by us:   tests/unit/core/test_task.py
	deleted by us:   tests/unit/providers/agents/builtin/test_responses_background.py

To fix up this pull request, you can check it out locally. See documentation: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/reviewing-changes-in-pull-requests/checking-out-pull-requests-locally