feat: OCR word_boxes fuer pixelgenaue Overlay-Positionierung

Backend: _ocr_cell_crop speichert jetzt word_boxes mit exakten
Tesseract/RapidOCR Wort-Koordinaten (left, top, width, height)
im Cell-Ergebnis. Absolute Bildkoordinaten, bereits zurueckgemappt.

Frontend: Slide-Hook nutzt word_boxes direkt wenn vorhanden —
jedes Wort wird exakt an seiner OCR-Position platziert. Kein
Pixel-Scanning noetig. Fallback auf alten Slide wenn keine Boxes.

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

klausur-service/backend/cv_cell_grid.py

@@ -221,6 +221,20 @@ def _ocr_cell_crop(
                     sum(w['conf'] for w in psm7_words) / len(psm7_words), 1
                 )
                 used_engine = 'cell_crop_v2_psm7'
+                # Remap PSM7 word positions back to original image coords
+                if up_w != cw or up_h != ch:
+                    sx = cw / max(up_w, 1)
+                    sy = ch / max(up_h, 1)
+                    for w in psm7_words:
+                        w['left'] = int(w['left'] * sx) + cx
+                        w['top'] = int(w['top'] * sy) + cy
+                        w['width'] = int(w['width'] * sx)
+                        w['height'] = int(w['height'] * sy)
+                else:
+                    for w in psm7_words:
+                        w['left'] += cx
+                        w['top'] += cy
+                words = psm7_words
 
     # --- Noise filter ---
     if text.strip():
@@ -235,6 +249,23 @@ def _ocr_cell_crop(
     result['text'] = text
     result['confidence'] = avg_conf
     result['ocr_engine'] = used_engine
+
+    # Store individual word bounding boxes (absolute image coordinates)
+    # for pixel-accurate overlay positioning in the frontend.
+    if words and text.strip():
+        result['word_boxes'] = [
+            {
+                'text': w.get('text', ''),
+                'left': w['left'],
+                'top': w['top'],
+                'width': w['width'],
+                'height': w['height'],
+                'conf': w.get('conf', 0),
+            }
+            for w in words
+            if w.get('text', '').strip()
+        ]
+
     return result