Verify `Agent.run_sync()` and `pydantic_ai.direct.model_request()` span coverage

`Agent.run_sync()` is not directly patched but likely works transitively since it calls `Agent.run()` internally. This should be verified with a test.

`pydantic_ai.direct.model_request()` is a standalone LLM call API that bypasses Agent entirely. The `Model.request` patch in `patches/model_request.py` may catch the underlying call, but this also needs verification.

Both should have explicit tests confirming span creation.