Test: Migrate the Eventarc tests off unittest.IsolatedAsyncioTestCase to pytest style

[AmaadMartin]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

N/A

2. Or, if no issue exists, describe the change:

Problem: Three Eventarc test classes inherited from unittest.IsolatedAsyncioTestCase. unittest awaits the coroutines of such a class itself, so pytest-asyncio never sees them. That makes @pytest.mark.asyncio, pytest.fixture and pytest.mark.parametrize inert inside those classes, and it hides the async intent from the pytest tooling the rest of the suite uses. The directory also disagreed with its own neighbour test_domain_specific_publish.py, which is already pytest style. A fourth module, test_config.py, was a plain unittest.TestCase, so the package advertised two styles at once.

Solution: I converted the three classes to plain pytest classes. setUp/tearDown in test_message_tool.py becomes an autouse fixture that patches through context managers. The old tearDown called mock.patch.stopall(), which drains unittest.mock's process-global registry and can therefore stop patches this class never started; the fixture unwinds only its own two patchers. I also converted test_config.py, which makes the whole package pytest style, and I dropped a dead os import from test_client.py in an import block this change already rewrites. This is a test-only change: src/google/adk/integrations/eventarc/** is untouched.

Collected-test count: 59 -> 70. The two self.subTest loops in test_message_tool.py each collected as one item. Their parametrized successors collect one item per case, which adds 8 invalid-input cases and 5 timestamp cases. I diffed --collect-only before and after: exactly 2 node IDs are removed and exactly 13 are added, and every added ID is an expansion of a former subTest case. No test is deleted, renamed, or weakened, and every class and method name is unchanged, so existing node IDs keep resolving.

Collision check. I listed the 100 open pull requests on this fork and inspected the adjacent ones. #179, #180 and #146 (async marker hygiene) touch no Eventarc file. #170 does touch test_message_tool.py: it rewrites the same setUp to use addCleanup, and it adds a TestMessageToolPatchHygiene(unittest.TestCase) guard. It does not migrate the class, so it is an overlap and not a duplicate. This change supersedes it, because the autouse fixture removes the mock.patch.stopall() hazard that #170 guards against. I branched from main rather than stacking, so the diff stays a pure migration. If #170 lands first, resolve the conflict by keeping this fixture and deleting the guard class: it cannot be instantiated once TestMessageTool is no longer a TestCase.

Testing Plan

Please describe the tests that you ran to verify your changes. This is required for all PRs that are not small documentation or typo fixes.

Unit Tests:

[x] I have added or updated unit tests for my change.
[x] All unit tests pass locally.

Commands run on the pushed commit, from the repository root:

uv run pytest tests/unittests/integrations/eventarc -q                        # 70 passed
uv run pytest tests/unittests/integrations/eventarc -q -o asyncio_mode=strict # 70 passed
uv run pytest tests/unittests/integrations/eventarc -q -n auto                # 70 passed

The baseline on main was 59 passed with 13 subtests passed. The strict run matters most: in strict mode pytest-asyncio does not await an async test that carries no marker. All 38 async tests pass there, so every marker is real. The run reports 0 skipped and no PytestUnhandledCoroutineWarning. The -n auto run matches the xdist setting that CI uses.

Formatters are a no-op afterwards:

uv run isort --check-only <the 3 files>   # clean
uv run pyink --check <the 3 files>        # 3 files would be left unchanged

Proof that the converted tests can fail. Coverage is not proof, so I broke the production source once per group and confirmed the tests catch it. I reverted each mutation immediately, and git diff main -- src/ is empty.

Mutation in src/	Tests that failed
Validation returns "status": "OK" instead of "ERROR"	all 8 new invalid_inputs cases: AssertionError: assert 'OK' == 'ERROR'
custom_attr["time"] pinned to a fixed timestamp	all 5 new time_valid_rfc3339 cases
custom_attr["time"] pinned to the first case's timestamp	4 of 5 cases; case 1 passes
Publisher client cache write removed	the 5 async tests in test_client.py; the sync test passes
_get_credential_id returns a constant	test_get_credential_id, which holds 25 converted asserts
Toolset drops its tool and skips client cleanup	the 2 async toolset tests and 1 sync test
Experimental warning text changed	the 2 assert any(...) warning tests
Success status literal changed	30 of 42 in test_message_tool.py; the 12 survivors are the error-path tests
All error status literals changed	the remaining 4 error-path tests
project_id default changed from None	test_config.py::test_valid_config
project_id widened to accept any type	test_config.py::test_invalid_config: DID NOT RAISE ValidationError

The third row is the one that justifies a deletion. The old loop called self.mock_eventarc_v1.reset_mock() on each iteration, and I removed it. Pinning the source to case 1's timestamp makes case 1 pass and cases 2 to 5 fail, which proves the mock does not leak between cases and the reset was dead code once the loop was gone.

Manual End-to-End (E2E) Tests:

Not applicable. No runtime code path changes, so there is nothing to exercise against a live service. The commands above are the deliverable.

Checklist

[x] I have read the CONTRIBUTING.md document.
[x] I have performed a self-review of my own code.
[x] I have commented my code, particularly in hard-to-understand areas.
[x] I have added tests that prove my fix is effective or that my feature works.
[x] New and existing unit tests pass locally with my changes.

CI note

Every test job passes: Unit Tests on Python 3.10 to 3.14, A2A v0.3 Tests on Python 3.10 to 3.14, and Mypy Check on Python 3.10 to 3.13.

Pre-commit Linter fails for an unrelated reason. The update-constraints hook regenerates constraints-3.*.txt with a newer --exclude-newer snapshot date, so the checkout is dirty and the gate rejects it. The failure names only those generated files, and this branch touches no file outside tests/unittests/integrations/eventarc/. The same job fails the same way on sibling pull requests that share no code with this one, for example #241 and #243. I did not fix it, because it is out of scope here.