test: run test_model_fields in CI again

[matthiasschabel]

tests/test_model_fields.py is skipped in all 15 jobs of the CI test matrix, so the pydantic and psygnal tests aren't running.

• The module imports BaseModel from pydantic_compat, which was never a declared test dependency. It came in through napari's app-model, which dropped it in 0.5.0.
• That import shares a try/except ImportError with import pydantic, so the module skips with "pydantic not installed" even though pydantic is installed.

This PR adds pydantic-compat to the test_thirdparty group and skips only when pydantic itself is missing, so a missing pydantic_compat now fails collection instead of skipping. The test_min job still skips the module.

I kept the shim rather than switching to pydantic.BaseModel, because test_pydantic_validate relies on it for .json() and the class-based Config, which warn under pydantic 2 and fail with filterwarnings = ["error"].

The 5 lines codecov/patch reports as missing on #155 are the pydantic serializer, which is only tested in this file.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.74%. Comparing base (87b7250) to head (ad579d1).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #164      +/-   ##
==========================================
+ Coverage   94.62%   95.74%   +1.12%     
==========================================
  Files         168      168              
  Lines        2231     2231              
==========================================
+ Hits         2111     2136      +25     
+ Misses        120       95      -25     

☔ View full report in Codecov by Harness.
📢 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.

[matthiasschabel]

@tlambert03 I don't understand the pydantic/CI stuff well enough to really carefully review this - it's mostly Claude's work reviewed by Codex...but it is pretty small, appears to fix the issue, and seems to make sense.