feat(halacha): NLI entailment validator via claude_session (#81.3) + task #86

#81.3 — a post-extraction validator that flags halachot whose rule_statement is
NOT entailed by its supporting_quote (the model over-reaching beyond its source).

- Engine: claude_session-as-judge (local CLI, zero API cost) per chaim's standing
  preference — one batched judge call per chunk, NOT a hosted NLI model.
- Pure, unit-tested helpers in halacha_quality: NLI_SYSTEM, build_nli_prompt,
  parse_nli_verdicts (fails OPEN — any shape/label ambiguity → 'entailed').
- halacha_extractor._nli_check wraps the call; fails OPEN on any error (e.g. no
  CLI in the container) so a flaky judge never blocks a genuine halacha.
- Non-entailed (neutral/contradiction) → quality_flag 'nli_unsupported' which
  blocks auto-approve (routes to pending_review) via the existing store gate.
- config: HALACHA_NLI_ENABLED/MODEL/EFFORT (effort 'low' — entailment is simple).

Verified: suite 166 passed (10 new); LIVE smoke test against the real claude CLI
returned ['entailed','neutral'] for a supported vs unsupported rule.

Also commits TaskMaster #86 (Nevo preamble/ratio: anti-contamination strip fix +
gold-set benchmark) capturing today's strip_nevo_preamble findings.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

mcp-server/src/legal_mcp/services/halacha_extractor.py

@@ -284,6 +284,27 @@ def _coerce_halacha(raw: dict, is_binding: bool = True) -> dict | None:
     }
 
 
+async def _nli_check(items: list[dict]) -> list[str]:
+    """Entailment verdict per item (rule ⊨ quote) via claude_session — #81.3.
+
+    Local CLI, zero cost. FAILS OPEN: any error returns all-'entailed' so a
+    flaky/unavailable judge (e.g. in the container) never blocks a halacha.
+    """
+    if not items:
+        return []
+    try:
+        raw = await claude_session.query_json(
+            halacha_quality.build_nli_prompt(items),
+            system=halacha_quality.NLI_SYSTEM,
+            model=config.HALACHA_NLI_MODEL or None,
+            effort=config.HALACHA_NLI_EFFORT or None,
+        )
+    except Exception as e:
+        logger.warning("halacha NLI check failed (fail-open, no flags): %s", e)
+        return ["entailed"] * len(items)
+    return halacha_quality.parse_nli_verdicts(raw, len(items))
+
+
 async def _extract_chunk(
     chunk_text: str,
     section_type: str,
@@ -511,6 +532,12 @@ async def _extract_impl(case_law_id: UUID, force: bool = False,
             if halacha_quality.FLAG_NON_DECISION in flags and coerced["rule_type"] != "obiter":
                 coerced["rule_type"] = "obiter"
             cleaned.append(coerced)
+        # #81.3 NLI entailment — one batched judge call per chunk (fail-open).
+        if config.HALACHA_NLI_ENABLED and cleaned:
+            verdicts = await _nli_check(cleaned)
+            for h, v in zip(cleaned, verdicts):
+                if v != "entailed" and halacha_quality.FLAG_NLI_UNSUPPORTED not in h["quality_flags"]:
+                    h["quality_flags"].append(halacha_quality.FLAG_NLI_UNSUPPORTED)
         if cleaned:
             embed_inputs = [
                 f"{h['rule_statement']} — {h['reasoning_summary']}".strip(" —")