[split-required] Split 500-850 LOC files (batch 2)

backend-lehrer (10 files):
- game/database.py (785 → 5), correction_api.py (683 → 4)
- classroom_engine/antizipation.py (676 → 5)
- llm_gateway schools/edu_search already done in prior batch

klausur-service (12 files):
- orientation_crop_api.py (694 → 5), pdf_export.py (677 → 4)
- zeugnis_crawler.py (676 → 5), grid_editor_api.py (671 → 5)
- eh_templates.py (658 → 5), mail/api.py (651 → 5)
- qdrant_service.py (638 → 5), training_api.py (625 → 4)

website (6 pages):
- middleware (696 → 8), mail (733 → 6), consent (628 → 8)
- compliance/risks (622 → 5), export (502 → 5), brandbook (629 → 7)

studio-v2 (3 components):
- B2BMigrationWizard (848 → 3), CleanupPanel (765 → 2)
- dashboard-experimental (739 → 2)

admin-lehrer (4 files):
- uebersetzungen (769 → 4), manager (670 → 2)
- ChunkBrowserQA (675 → 6), dsfa/page (674 → 5)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

klausur-service/backend/qdrant_nibis.py

@@ -0,0 +1,79 @@
+"""
+Qdrant Vector Database Service — NiBiS RAG Search for Klausurkorrektur.
+"""
+
+from typing import List, Dict, Optional
+from qdrant_client.models import Filter, FieldCondition, MatchValue
+
+from qdrant_core import get_qdrant_client
+
+
+async def search_nibis_eh(
+    query_embedding: List[float],
+    year: Optional[int] = None,
+    subject: Optional[str] = None,
+    niveau: Optional[str] = None,
+    limit: int = 5
+) -> List[Dict]:
+    """
+    Search in NiBiS Erwartungshorizonte (public, pre-indexed data).
+
+    Unlike search_eh(), this searches in the public NiBiS collection
+    and returns plaintext (not encrypted).
+
+    Args:
+        query_embedding: Query vector
+        year: Optional year filter (2016, 2017, 2024, 2025)
+        subject: Optional subject filter
+        niveau: Optional niveau filter (eA, gA)
+        limit: Max results
+
+    Returns:
+        List of matching chunks with metadata
+    """
+    client = get_qdrant_client()
+    collection = "bp_nibis_eh"
+
+    # Build filter
+    must_conditions = []
+
+    if year:
+        must_conditions.append(
+            FieldCondition(key="year", match=MatchValue(value=year))
+        )
+    if subject:
+        must_conditions.append(
+            FieldCondition(key="subject", match=MatchValue(value=subject))
+        )
+    if niveau:
+        must_conditions.append(
+            FieldCondition(key="niveau", match=MatchValue(value=niveau))
+        )
+
+    query_filter = Filter(must=must_conditions) if must_conditions else None
+
+    try:
+        results = client.search(
+            collection_name=collection,
+            query_vector=query_embedding,
+            query_filter=query_filter,
+            limit=limit
+        )
+
+        return [
+            {
+                "id": str(r.id),
+                "score": r.score,
+                "text": r.payload.get("text", ""),
+                "year": r.payload.get("year"),
+                "subject": r.payload.get("subject"),
+                "niveau": r.payload.get("niveau"),
+                "task_number": r.payload.get("task_number"),
+                "doc_type": r.payload.get("doc_type"),
+                "variant": r.payload.get("variant"),
+            }
+            for r in results
+        ]
+    except Exception as e:
+        print(f"NiBiS search error: {e}")
+        return []