Suppress expected warnings in tests

[TheJulianJES]

This filters expected warnings triggered by various tests.
Alternatively, we can also add this to our pyproject.toml. Adding it to the individual tests or modules doesn't really work here, except if we want to add it to basically all.

[TheJulianJES]

At the moment, unique IDs only being unique within a platform is expected and this is logged quite a bit.
I can remove this, of corse, but I think it might make sense to keep until we've migrated away from intentionally using that?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.51%. Comparing base (9d03d63) to head (8f9e83f).
⚠️ Report is 31 commits behind head on dev.

Additional details and impacted files

@@           Coverage Diff           @@
##              dev     #696   +/-   ##
=======================================
  Coverage   97.51%   97.51%           
=======================================
  Files          62       62           
  Lines       10949    10949           
=======================================
  Hits        10677    10677           
  Misses        272      272           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[zigpy-review-bot]

Skimmed both suppressed warnings against their emission sites; both are legitimate "this is how the tests are intentionally constructed" noise, not real bugs being silenced.

• is_manufacturer_specific without manufacturer_code is emitted by zigpy at zigpy/zcl/__init__.py:612 when an attribute lookup matches a manufacturer-specific attr that declares no explicit manufacturer code. In the suite this fires from minimal test fixtures like FakeManufacturerCluster in tests/test_button.py:186-188 and the FrameControl(is_manufacturer_specific=manufacturer is not None, …) construction in tests/common.py:174-184. Those fixtures are correct-as-is for what they test; the warning is zigpy nagging production callers to be explicit. Suppression is fine.
• Unique IDs are unique only with platform prefix is emitted by ZHA's own test code (tests/test_discover.py:730-737) and represents acknowledged tech debt (per your inline comment). Suppression is also fine here.

Few notes:

1. Scoping is good. Adding via pytest.mark.filterwarnings in pytest_collection_modifyitems keeps the suppression visible at collection time and leaves the door open for individual tests to override with @pytest.mark.filterwarnings("error::DeprecationWarning") if they ever want to assert the warning does fire. Strictly preferable to a blanket filterwarnings in [tool.pytest.ini_options] — you weighed that alternative in the PR body and landed in the right place IMO.
2. Cross-PR re: #695. Running the suite on this branch still shows ~50 pytest-asyncio 0.26 asyncio.get_event_loop_policy is deprecated warnings (Python 3.14). #696 deliberately does not silence those, which is the right call — #695 (bump pytest-asyncio to 1.3.0) fixes them at source. Worth landing #695 first if anything, since once it's in #696's signal-to-noise improvement will be much more dramatic.
3. Optional nit. The two filter strings are quite verbose. Since they're plain prefixes (no real regex semantics needed in the body), re.escape("Attribute '") + ... style helpers or just shorter prefixes (r"ignore:Attribute .* has .is_manufacturer_specific.:DeprecationWarning") would scan more easily. Strictly cosmetic.
4. Nothing leaks into runtime. conftest.py only applies markers during collection, no production-code touchpoints.

CI is green on 3.12 / 3.13 / pre-commit. LGTM once you take it out of draft (after #695 if you want the cleaner before/after).