[M5] Formalize root-cause label schema

[askmy-stack]

Summary

• Add a versioned RootCauseCode / RootCauseLabel Pydantic schema in runtime/schemas/root_cause.py, replacing the ad hoc free-text root_cause strings diagnosis previously produced.
• RootCauseLabel pairs a RootCauseCode with its FailureCategory and a schema_version, validating that the two stay consistent.
• GroundTruthBlock.root_cause (case YAML) is now typed as RootCauseCode, so CaseRegistry.from_yaml_file / from_directory reject unknown root-cause codes at load time via BenchmarkCaseModel.model_validate.
• BenchmarkCase.root_cause_label exposes the resolved, categorized label for a loaded case.

Scope

• Pydantic model for root-cause labels ✅ (runtime/schemas/root_cause.py)
• Align case YAML ground_truth.root_cause ✅ (existing tool-semantic-drift-001.yaml value already matches RootCauseCode.PLAN_IDENTIFIER_SEMANTICS_CHANGED, now enforced)

Acceptance criteria

• Schema in runtime/schemas/
• Validator used by CaseRegistry

Fixes #27

Test plan

• make ci (ruff check, ruff format --check, mypy, pytest) passes locally: 18 tests pass, 0 mypy errors.
• New tests/unit/test_root_cause.py covers: category inference from code, accepting raw strings, rejecting a mismatched category, rejecting unknown codes, CaseRegistry successfully validating the existing case file, and CaseRegistry raising ValidationError for a case file with an unrecognized root_cause.
• Existing test suite (including tests/unit/test_detectors.py's diagnosis.root_cause == "plan_identifier_semantics_changed" assertion) still passes unchanged since RootCauseCode is a StrEnum.

Made with Cursor