feat(ontology): add PGE Evaluator stage to owl-generator

[FiifiB]

Summary

Turns OWL ontology generation into a real Planner→Generator→Evaluator (PGE)
loop. Today agent_owl_generator does single-shot generation plus a pitfall-tool
fix loop, but has no deterministic Evaluator — so hard structural defects can
survive into the delivered ontology. This PR adds a Stage-1 evaluator that scores
the generated ontology against the source metadata and feeds concrete retry-hints
back to the generator on Tier-1 defects, bounded by a hard cap.

This is the first of two independent PGE PRs (ontology generation; entity/
relationship mapping). It is additive and self-contained.

What changed

• agent_owl_generator/engine.py
  • _evaluate_ontology_stage() — parses the Turtle, runs deterministic Tier-1
    checks (orphan classes, dangling rdfs:domain/rdfs:range, naming violations,
    duplicate classes) and returns a retry-hint on hard defects. Fails open —
    any parse/dependency error returns None, so the evaluator never blocks
    delivery.
  • Evaluator wired into the loop after the pitfall loop, bounded by
    MAX_OWL_EVAL_ROUNDS, and only retries while an iteration remains so a usable
    ontology is never discarded by exhausting MAX_ITERATIONS.
  • MAX_OUTPUT_TOKENS = 16000 so exhaustive attribute coverage isn't silently
    truncated past the old 4096 ceiling.
  • Stronger prompt: # ATTRIBUTE COVERAGE section + get_table_detail-per-table
    workflow step → exhaustive (not curated) datatype-property coverage.
• New agents/pge_eval/ slice (gold-free, intrinsic, computed from the
  ontology + source schema only): normalize.py + ontology_metrics.evaluate_ontology.
  The package root is intentionally minimal; the full scorecard/CLI lands in a
  separate change, so importers depend on the concrete submodule.

Testing

uv run pytest tests/units/pge_eval/test_ontology_metrics.py tests/units/pge_eval/test_owl_evaluator_stage.py tests/units/ontology/test_owl_generator.py -q → 39 passed.

Broader sanity: tests/units/{agents,ontology,pge_eval} → 565 passed, 11 skipped.

This pull request and its description were written by Isaac.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Fiifi Botchway seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}