fix(agents): Impressum+Cookie delegieren MC-Laden ans Main Tool — Scope-Filter + Maßnahmen

Regression: Der v3-Agent-Pfad baute eine parallele MC-Pipeline
(_load_impressum_mcs / _load_cookie_mcs, Roh-SELECT) und lief damit an
allen Schutzmechanismen der Engine vorbei → GOV/Branchen-MCs als HIGH bei
OEM/Zulieferer, fremde MCs (Bestellbestätigung), und action=check_question
(Fragen statt Maßnahmen im Frontend).

- Agent delegiert MC-Laden an rag_document_checker._load_controls
  (P72-Scope, check_type='text', fits_doc_type/scope_requires).
- Subtraktives Sektor-Gate (SECTOR_PREFIXES) + Themen-Gate am Agent-Rand.
- action = konkrete Maßnahme (Imperativ) statt check_question.
- rag_document_checker: from __future__ import annotations (3.9-Import).
- mcs: Name-Pattern erkennt "Aktiengesellschaft" (OEM-Impressums).
- Tote GT-/Semantic-/Routes-Tests wiederbelebt (v3-Mismatch +
  agent.cascade-Patch-Target). Alle 72 Specialist-Tests grün.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend-compliance/tests/test_impressum_groundtruth.py

@@ -11,7 +11,10 @@ import asyncio
 import pytest
 
 from compliance.services.specialist_agents import AgentInput, ImpressumAgent
-from tests.fixtures.impressum_groundtruth import ALL_GROUND_TRUTH
+from tests.fixtures.impressum_groundtruth import (
+    ALL_GROUND_TRUTH,
+    CORE_IMPRESSUM_MCS,
+)
 
 
 def _run(coro):
@@ -19,15 +22,40 @@ def _run(coro):
 
 
 @pytest.fixture(autouse=True)
-def _no_llm(monkeypatch):
-    """Skip LLM-Eskalation in den GT-Tests — wir testen MC-Pattern,
-    nicht LLM-Halluzinationen."""
-    async def _no_cascade(*a, **kw): return None, []
+def _agent_offline(monkeypatch):
+    """GT-Tests offline + deterministisch: LLM-Eskalation aus, MC-Laden ueber
+    die gemockte Main-Tool-Engine (CORE_IMPRESSUM_MCS), Embedding aus. Der
+    Agent delegiert jetzt ans Main Tool — daher `_load_controls` mocken."""
+    import copy
+
+    async def _no_cascade(*a, **kw):
+        return None, []
     monkeypatch.setattr(
-        "compliance.services.specialist_agents.impressum.agent.cascade",
+        "compliance.services.specialist_agents._semantic_validator.cascade",
         _no_cascade,
     )
 
+    async def _fake_load(doc_type, db_url, limit, business_scope=None):
+        return copy.deepcopy(CORE_IMPRESSUM_MCS)
+    monkeypatch.setattr(
+        "compliance.services.rag_document_checker._load_controls",
+        _fake_load,
+    )
+
+    async def _no_embed(*a, **kw):
+        return None
+
+    async def _no_match(*a, **kw):
+        return set()
+    monkeypatch.setattr(
+        "compliance.services.mc_embedding_matcher.ensure_mc_embeddings",
+        _no_embed, raising=False,
+    )
+    monkeypatch.setattr(
+        "compliance.services.mc_embedding_matcher.embedding_match",
+        _no_match, raising=False,
+    )
+
 
 @pytest.mark.parametrize("gt", ALL_GROUND_TRUTH, ids=lambda g: g.name)
 def test_no_false_positives_on_expected_clean(gt):