fix(util-genai): keep multimodal tests Python 3.8-compatible

[sipercai]

Description

Keep the synchronized multimodal tests runnable for downstream Python 3.8
consumers without changing LoongSuite runtime behavior or its declared
requires-python >=3.9 support floor.

This replaces parenthesized multi-context with statements with nested context
managers and replaces runtime-evaluated built-in generic annotations with
typing.Tuple / typing.Dict in the affected tests. No production source files
are changed.

Fixes # (N/A)

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality not to work as expected)
• This change requires a documentation update

How Has This Been Tested?

• Python 3.8 compileall across util/opentelemetry-util-genai/src and tests
• Python 3.8 full multimodal test directory: 135 passed, 1 credential-dependent test skipped
• tox -e py39-test-util-genai: 330 passed, 1 credential-dependent test skipped
• tox -e py312-test-util-genai: 330 passed, 1 credential-dependent test skipped
• tox -e lint-util-genai: pylint 10.00/10
• tox -e typecheck: 0 errors, 0 warnings
• Fresh-cache tox -e precommit

Does This PR Require a Core Repo Change?

• Yes. - Link to PR:
• No.

Checklist:

• Followed the style guidelines of this project
• Changelogs have been updated
• Unit tests have been added
• Documentation has been updated

Validation Evidence

Spec and Scope

• Linked issue/spec: N/A
• Approved scope: retain only downstream Python 3.8 test compatibility; do not change uploader retirement behavior
• Changed surface: util-genai tests and CHANGELOG-loongsuite.md
• Runtime source: unchanged from origin/main

Local Checks

Check	Command	Result	Notes
Static readiness	python3 "$PIPELINE_SKILL_DIR/scripts/check_loongsuite_pr_readiness.py" --repo .	pass	Expected warning: no instrumentation plugin directory changed
Precommit	PRE_COMMIT_HOME="$(mktemp -d)" tox -e precommit	pass	Fresh hook cache; clean worktree afterward
Python 3.8 syntax	python3.8 -m compileall -q util/opentelemetry-util-genai/src util/opentelemetry-util-genai/tests	pass	Covers all util-genai source and test files
Python 3.8 focused tests	PYTHONPATH="$PWD/util/opentelemetry-util-genai/src" python3.8 -m pytest -q util/opentelemetry-util-genai/tests/_multimodal_upload	pass	135 passed, 1 credential-dependent test skipped
Python 3.9 tests	tox -e py39-test-util-genai	pass	330 passed, 1 credential-dependent test skipped
Python 3.12 tests	tox -e py312-test-util-genai	pass	330 passed, 1 credential-dependent test skipped
Lint	tox -e lint-util-genai	pass	pylint 10.00/10
Typecheck	tox -e typecheck	pass	0 errors, 0 warnings
Independent diff review	scoped branch review	pass	Zero actionable findings
Privacy scan	branch-diff path, identity, credential, internal-URL, and private-artifact scan	pass	No matches

Fix Verification

• On the synchronized baseline, Python 3.8 cannot parse the parenthesized
  multi-context with statements and evaluates tuple[...] / dict[...]
  annotations during test collection.
• On this branch, Python 3.8 compileall passes and the complete multimodal test
  directory collects and runs successfully.
• Python 3.9 and Python 3.12 full util-genai suites remain green.

Business Isolation

Scenario	Status	Evidence
prepare/start fault	N/A	Test-only syntax and annotation changes
response mapping/stop fault	N/A	No runtime callback changed
fail callback fault	N/A	No runtime callback changed
stream chunk/final callback fault	N/A	No streaming code changed
early close/cancellation/GeneratorExit	N/A	No generator lifecycle changed
cross-task or cross-Context stream	N/A	No Context handling changed
clean sibling after fault	N/A	No runtime state changed

Real E2E Matrix

Scenario	Status	Command or Demo	Evidence
non-streaming	N/A	Test-only change	No model/provider path changed
streaming	N/A	Test-only change	No stream path changed
concurrency	N/A	Test-only change	No runtime concurrency changed
agent/tool/ReAct	N/A	Test-only change	No agent/tool path changed
tool-heavy	N/A	Test-only change	No tool path changed
error path	N/A	Test-only change	No runtime error path changed

Telemetry and Weaver

Check	Status	Command or Artifact	Notes
Span tree / span kinds	N/A	Test-only change	No telemetry output changed
Content capture modes	N/A	Test-only change	No content-capture behavior changed
Concurrency isolation	N/A	Test-only change	No trace or Context state changed
Weaver live-check	N/A	Test-only change	No Weaver-checkable telemetry delta

CI

• Current published head: LoongSuite package tests, util-genai lint, precommit,
  changelog, and license checks passed.
• Superseded failure: typecheck failed on the retirement-worker implementation
  that this update removes; local typecheck passes on the narrowed head.
• Follow-up needed: rerun GitHub checks after pushing this update, then
  synchronize the compatibility commit into the downstream Python 3.8 branch.

[ralf0131]

Summary

Replaces per-pair daemon thread creation with a single lazily-started FIFO retirement worker for multimodal uploader cleanup. This prevents unbounded thread growth during rapid runtime configuration changes. Also fixes Python 3.8 test compatibility by replacing parenthesized context managers and lowercase generic types with their typing module equivalents.

Overall: LGTM. The approach is clean, the double-check locking pattern is correct, fork safety is well-handled, and the regression tests comprehensively cover the new behavior (single-worker reuse, FIFO ordering, fail-open on shutdown failure, fork reset, atexit sentinel).

Findings

• [Info] multimodal_upload_hook.py:51-54 — The globals().get("_retired_pair_at_fork_registered", False) pattern is slightly unusual for a module-level guard. A plain _retired_pair_at_fork_registered = False at module level would be equivalent and more readable. The current form works correctly but may confuse readers into thinking there is a circular dependency being avoided. Non-blocking.
• [Info] multimodal_upload_hook.py:155-160 — _reset_retired_pair_worker_after_fork intentionally does not reset _retired_pair_at_fork_registered, which is correct since os.register_at_fork handlers are inherited by the child process. A brief inline comment explaining this design choice would help future maintainers.

Suggestions

• Consider adding a one-line comment above the _retired_pair_at_fork_registered module variable explaining that it survives fork resets because the register_at_fork handler is inherited by child processes.

---

Automated review by github-manager-bot

[ralf0131]

Summary

LGTM — This PR correctly fixes Python 3.8 compatibility issues in the multimodal test files by replacing Python 3.9+ type hints (tuple[str, int] → Tuple[str, int]) and Python 3.10+ parenthesized context managers with nested with statements.

The changes are minimal, focused, and do not affect production code. They ensure downstream consumers on Python 3.8 can run the tests without issues.

---

Automated review by github-manager-bot

[ralf0131]

LGTM. Clean mechanical fix — replaces Python 3.9+ built-in generics (tuple[], dict[]) with typing.Tuple/typing.Dict and parenthesized multi-context with with nested context managers in test files only. No production code changes.

---

Automated review by github-manager-bot