feat(sources): Review-Queue-UI fuer LLM-Klassifikations-Vorschlaege (Admin)

- Tab-Schalter im Quellen-Modal: "Quellenliste" vs. "Klassifikations-Review"
  (Review-Tab nur fuer org_admin sichtbar, mit Pending-Counter-Badge).
- Review-Karten zeigen Diff aktueller Wert -> LLM-Vorschlag pro Achse,
  Konfidenz-Indikator (gruen/gelb/rot), LLM-Begruendung, Buttons fuer
  Uebernehmen / Verwerfen / Neu klassifizieren.
- Toolbar: Konfidenz-Filter, "Klassifikation starten" (Bulk im Hintergrund),
  "Alle >= 0.85 genehmigen" (Bulk-Approve).
- API-Wrapper in api.js fuer alle 6 neuen Endpoints + erweiterte listSources-Filter.
- Backend-Endpoint POST /api/sources/classification/bulk-approve (Admin-only).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/routers/sources.py

@@ -937,3 +937,62 @@ async def trigger_bulk_classify(
         raise HTTPException(status_code=400, detail="limit muss zwischen 1 und 500 liegen")
     background_tasks.add_task(_bulk_classify_background, limit, only_unclassified)
     return {"status": "started", "limit": limit, "only_unclassified": only_unclassified}
+
+
+@router.post("/classification/bulk-approve")
+async def bulk_approve_classifications(
+    min_confidence: float = 0.85,
+    current_user: dict = Depends(get_current_user),
+    db: aiosqlite.Connection = Depends(db_dependency),
+):
+    """Genehmigt alle Pending-Vorschlaege ueber dem confidence-Schwellwert (nur Admins).
+
+    Globale Quellen werden nur bearbeitet, wenn der Aufrufer org_admin ist;
+    Tenant-eigene Quellen sowieso.
+    """
+    if current_user.get("role") != "org_admin":
+        raise HTTPException(status_code=403, detail="Nur Admins koennen Bulk-Approve nutzen")
+    tenant_id = current_user.get("tenant_id")
+    cursor = await db.execute(
+        """SELECT id, proposed_political_orientation, proposed_media_type,
+                  proposed_reliability, proposed_state_affiliated,
+                  proposed_country_code, proposed_alignments_json, tenant_id
+           FROM sources
+           WHERE proposed_political_orientation IS NOT NULL
+             AND COALESCE(proposed_confidence, 0) >= ?
+             AND (tenant_id IS NULL OR tenant_id = ?)""",
+        (min_confidence, tenant_id),
+    )
+    rows = [dict(r) for r in await cursor.fetchall()]
+    approved_ids: list[int] = []
+    for src in rows:
+        try:
+            proposed_aligns = json.loads(src.get("proposed_alignments_json") or "[]")
+        except (json.JSONDecodeError, TypeError):
+            proposed_aligns = []
+        await db.execute(
+            """UPDATE sources SET
+                political_orientation = ?,
+                media_type = ?,
+                reliability = ?,
+                state_affiliated = ?,
+                country_code = ?,
+                classification_source = 'llm_approved',
+                classified_at = CURRENT_TIMESTAMP
+            WHERE id = ?""",
+            (
+                src["proposed_political_orientation"],
+                src["proposed_media_type"],
+                src["proposed_reliability"],
+                1 if src.get("proposed_state_affiliated") else 0,
+                src.get("proposed_country_code"),
+                src["id"],
+            ),
+        )
+        await _replace_alignments(
+            db, src["id"], [a for a in proposed_aligns if a in ALLOWED_ALIGNMENTS]
+        )
+        await _clear_proposed(db, src["id"])
+        approved_ids.append(src["id"])
+    await db.commit()
+    return {"approved_count": len(approved_ids), "min_confidence": min_confidence}