feat: use local RapidOCR as default in ocr_region_paddle(), remote as fallback

RapidOCR uses the same PP-OCRv5 ONNX models locally, avoiding 504 timeouts
from remote PaddleOCR on large images. Set FORCE_REMOTE_PADDLE=1 to bypass.

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

klausur-service/backend/cv_ocr_engines.py

@@ -7,6 +7,7 @@ DATENSCHUTZ: Alle Verarbeitung erfolgt lokal.
 
 import io
 import logging
+import os
 import re
 from typing import Any, Dict, List, Optional, Tuple
 
@@ -392,11 +393,32 @@ async def ocr_region_paddle(
     img_bgr: np.ndarray,
     region: Optional["PageRegion"] = None,
 ) -> List[Dict[str, Any]]:
-    """Run OCR via remote PaddleOCR service (Hetzner).
+    """Run OCR via local RapidOCR (default) or remote PaddleOCR (fallback).
 
-    If *region* is given, crops before sending. Otherwise sends the full image.
+    Uses RapidOCR (same PP-OCRv5 ONNX models) locally for speed and reliability.
-    Returns word dicts in the standard format (left/top in absolute coords).
+    Falls back to remote PaddleOCR service only if:
+    - env FORCE_REMOTE_PADDLE=1 is set, or
+    - RapidOCR fails or returns no words
     """
+    force_remote = os.environ.get("FORCE_REMOTE_PADDLE", "").strip() == "1"
+
+    if not force_remote:
+        try:
+            if region is None:
+                h, w = img_bgr.shape[:2]
+                _region = PageRegion(type="full_page", x=0, y=0, width=w, height=h)
+            else:
+                _region = region
+
+            words = ocr_region_rapid(img_bgr, _region)
+            if words:
+                logger.info("ocr_region_paddle: used local RapidOCR (%d words)", len(words))
+                return words
+            logger.warning("ocr_region_paddle: RapidOCR returned 0 words, trying remote")
+        except Exception as e:
+            logger.warning("ocr_region_paddle: RapidOCR failed (%s), trying remote", e)
+
+    # --- Remote PaddleOCR fallback (Hetzner x86_64) ---
     from services.paddleocr_remote import ocr_remote_paddle
 
     if region is not None:
@@ -431,6 +453,7 @@ async def ocr_region_paddle(
         return []
 
     words, _w, _h = await ocr_remote_paddle(jpg_buf.tobytes(), filename="scan.jpg")
+    logger.info("ocr_region_paddle: used remote PaddleOCR (%d words)", len(words))
 
     # Scale coordinates back to original size and shift to absolute image space
     inv_scale = 1.0 / scale if scale != 1.0 else 1.0