fix(tests): guard OpenAI embedding tests and drop read-only property monkeypatch (#288)

[Diyaaa-12]

Extends the skipif guard on TestOpenAIBatchSplitting to require both pydantic_ai and openai, and removes a monkeypatch on a read-only property in the classification stub-fallback fixture. Both were latent bugs discovered during #287's CI review, tracked here per your request.

Related Issue

• Fixes tests: latent failures in envs with pydantic-ai installed (embedding openai guard + classification settings monkeypatch) #288

Change Summary

• Extended the skipif guard on TestOpenAIBatchSplitting (test_pydantic_ai_embedding_adapter.py) to require both pydantic_ai and openai, since _build_model imports pydantic_ai.embeddings.openai which needs the openai package.
• Removed monkeypatch.setattr(settings, "llm_model_name", None) in the classification stub-fallback fixture — it's a read-only derived property that already resolves to None once llm_provider/llm_model are patched.

Type of Change

• feat: New feature
• fix: Bug fix
• refactor: Code restructuring
• docs: Documentation
• chore: Build/tooling
• test: Tests
• ci: CI/CD
• perf: Performance
• style: Code style

Checklist

• Architecture rules followed (no Domain -> Infrastructure imports)
• Tests pass
• Linting passes (ruff check src/)

How to Test

uv sync --extra pydantic-ai
uv run pytest tests/ -v --tb=short
# affected tests pass or skip cleanly

uv sync --extra pydantic-ai --extra openai
uv run pytest tests/unit/_core/infrastructure/embedding/test_pydantic_ai_embedding_adapter.py -v --tb=short
# TestOpenAIBatchSplitting (4 tests) run and pass

Scope kept to just these two test fixes as requested — the CI-matrix decision (whether the test job should install --extra pydantic-ai/openai) is left out of scope.

---

Governor Footer

• trigger: yes
• reviewer: self-structured
• rounds: 1
• r-points-fixed: 0
• r-points-deferred: 0
• r-points-rejected: 0
• touched-adr-consequences: none
• pr-scope-notes: test-only fix scoped to the two latent bugs from tests: latent failures in envs with pydantic-ai installed (embedding openai guard + classification settings monkeypatch) #288; CI-matrix extras decision intentionally out of scope
• final-verdict: merge-ready
• links: fix(tests): guard OpenAI embedding tests and drop read-only property monkeypatch (#288) #291

[Mr-DooSun]

Thanks for the careful fix, @Diyaaa-12 — and for splitting these two latent failures out of the #287 review into their own tracked issue.

Reviewed against the source and confirmed both fixes:

• Bug 1 (embedding guard): Requiring openai in the skip guard is exactly right. PydanticAIEmbeddingAdapter.__init__ → _build_model runs from pydantic_ai.embeddings.openai import OpenAIEmbeddingModel when the test builds an openai: config with an api_key, so TestOpenAIBatchSplitting genuinely needs openai, not just pydantic_ai.
• Bug 2 (read-only property): Confirmed Settings.llm_model_name is a setter-less @property derived from llm_provider/llm_model, so monkeypatch.setattr(..., "llm_model_name", None) raised AttributeError at setup (reproduced locally). The two remaining lines already drive the property to None, and the container's LLM selector reads that derived value at resolution time (core_container.py:56), so test_core_container_llm_model_is_test_model still resolves to TestModel. Behavior-preserving.

Non-blocking note: the new guard checks pydantic_ai + openai but not tiktoken (needed by _init_tiktoken). No real gap — both the openai and pydantic-ai extras pull in tiktoken, so whenever openai resolves it's present. Fine to leave as-is.

Also agree with keeping the CI-matrix decision (whether the test job installs these extras) out of scope — worth remembering that until that lands, these two tests stay skipped in CI.

Approving — I'll merge shortly. Thanks again for the clean, well-documented contribution!