fix(ocr-pipeline): configure RapidOCR for German + tighter word detection

- Switch to PP-OCRv5 Latin model (supports ä, ö, ü, ß)
- Use SERVER model for better accuracy
- Lower Det.unclip_ratio 1.6→1.3 to reduce word merging
- Raise Det.box_thresh 0.5→0.6 for stricter detection

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

klausur-service/backend/cv_vocab_pipeline.py

@@ -2196,6 +2196,7 @@ RAPIDOCR_AVAILABLE = False
 
 try:
     from rapidocr import RapidOCR as _RapidOCRClass
+    from rapidocr import LangRec as _LangRec, OCRVersion as _OCRVersion, ModelType as _ModelType
     RAPIDOCR_AVAILABLE = True
     logger.info("RapidOCR available — can be used as alternative to Tesseract")
 except ImportError:
@@ -2203,11 +2204,21 @@ except ImportError:
 
 
 def _get_rapid_engine():
-    """Lazy-init RapidOCR engine (downloads models on first use)."""
+    """Lazy-init RapidOCR engine with PP-OCRv5 Latin model for German support."""
     global _rapid_engine
     if _rapid_engine is None:
-        _rapid_engine = _RapidOCRClass()
+        _rapid_engine = _RapidOCRClass(params={
-        logger.info("RapidOCR engine initialized")
+            # PP-OCRv5 Latin model — supports German umlauts (ä, ö, ü, ß)
+            "Rec.lang_type": _LangRec.LATIN,
+            "Rec.model_type": _ModelType.SERVER,
+            "Rec.ocr_version": _OCRVersion.PPOCRV5,
+            # Tighter detection boxes to reduce word merging
+            "Det.unclip_ratio": 1.3,
+            "Det.box_thresh": 0.6,
+            # Silence verbose logging
+            "Global.log_level": "critical",
+        })
+        logger.info("RapidOCR engine initialized (PP-OCRv5 Latin, unclip_ratio=1.3)")
     return _rapid_engine