feat(principles): decision-level panel extraction regime — cap-5 + dedup-frees-slot (Phase B, #152)

extract() routes to _extract_via_panel when HALACHA_PANEL_REGIME_ENABLED: the
3-model panel proposes → votes/score → approval rule → dedup vs corpus (known
links as citation, frees a cap slot) → cap HALACHA_PANEL_MAX_NEW genuinely-new
principles/decision (by score), rest dropped. Replaces single-model auto-approve;
legacy path kept as <2-judge fallback. db.store_panel_principles persists the
pre-decided verdict + source-aware canonical create/link (G9 reviewer=panel:...).
Dry-run validated on 29468-08-23: ~18 → 4 principles. 6 new tests; full suite 422 green.

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

mcp-server/src/legal_mcp/config.py

@@ -179,6 +179,10 @@ HALACHA_PANEL_MAX_NEW = int(os.environ.get("HALACHA_PANEL_MAX_NEW", "5"))
 # a floor misses genuine cross-model agreement → undercounts votes → over-culls.
 # Calibrate against the gold-set in Phase C before the production cull.
 HALACHA_PANEL_MATCH_COSINE = float(os.environ.get("HALACHA_PANEL_MATCH_COSINE", "0.80"))
+# When on (default), extraction uses the decision-level 3-model panel regime above
+# instead of the legacy per-chunk single-model auto-approve. Set false to fall back
+# to the legacy path (e.g. if all three judges are unreachable).
+HALACHA_PANEL_REGIME_ENABLED = os.environ.get("HALACHA_PANEL_REGIME_ENABLED", "true").lower() == "true"
 
 # Halacha dedup-on-insert — within-precedent semantic cosine ceiling. Before
 # storing a halacha, store_halachot_for_chunk skips it if its rule-embedding has