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/tests/test_halacha_quality.py

@@ -91,3 +91,58 @@ def test_flags_accumulate():
 
 def test_normalize_text_quote_variants():
     assert hq.normalize_text('עע"מ   317/10') == hq.normalize_text("עע״מ 317/10")
+
+
+# ── #81.3 NLI entailment — pure prompt + parser ──
+
+def test_build_nli_prompt_contains_pairs():
+    items = [
+        {"rule_statement": "כלל אלף", "supporting_quote": "ציטוט אלף"},
+        {"rule_statement": "כלל בית", "supporting_quote": "ציטוט בית"},
+    ]
+    p = hq.build_nli_prompt(items)
+    assert "כלל אלף" in p and "ציטוט בית" in p
+    assert "זוג 1" in p and "זוג 2" in p
+
+
+@pytest.mark.parametrize("raw,n,expected", [
+    (["entailed", "neutral"], 2, ["entailed", "neutral"]),
+    (["ENTAILED", "Contradiction"], 2, ["entailed", "contradiction"]),  # case-insensitive
+    ([{"verdict": "neutral"}, {"verdict": "entailed"}], 2, ["neutral", "entailed"]),  # dict shape
+    (["entailed"], 2, ["entailed", "entailed"]),          # length mismatch -> fail-open
+    (None, 2, ["entailed", "entailed"]),                  # non-list -> fail-open
+    (["bananas", "neutral"], 2, ["entailed", "neutral"]), # unknown label -> entailed
+])
+def test_parse_nli_verdicts(raw, n, expected):
+    assert hq.parse_nli_verdicts(raw, n) == expected
+
+
+# ── _nli_check (async, via claude_session) — fail-open + verdict mapping ──
+
+def test_nli_check_fail_open(monkeypatch):
+    import asyncio
+    from legal_mcp.services import halacha_extractor as he
+
+    async def boom(*a, **k):
+        raise RuntimeError("no claude CLI here")
+    monkeypatch.setattr(he.claude_session, "query_json", boom)
+    items = [{"rule_statement": "a", "supporting_quote": "b"}]
+    assert asyncio.run(he._nli_check(items)) == ["entailed"]  # never blocks
+
+
+def test_nli_check_maps_verdicts(monkeypatch):
+    import asyncio
+    from legal_mcp.services import halacha_extractor as he
+
+    async def fake(*a, **k):
+        return ["entailed", "neutral"]
+    monkeypatch.setattr(he.claude_session, "query_json", fake)
+    items = [{"rule_statement": "a", "supporting_quote": "b"},
+             {"rule_statement": "c", "supporting_quote": "d"}]
+    assert asyncio.run(he._nli_check(items)) == ["entailed", "neutral"]
+
+
+def test_nli_check_empty():
+    import asyncio
+    from legal_mcp.services import halacha_extractor as he
+    assert asyncio.run(he._nli_check([])) == []