fix(ocr-pipeline): improve page crop spine detection and cell assignment

1. page_crop: Score all dark runs by center-proximity × darkness ×
   narrowness instead of picking the widest. Fixes ad810209 where a
   wide dark area at 35% was chosen over the actual spine at 50%.

2. cv_words_first: Replace x-center-only word→column assignment with
   overlap-based three-pass strategy (overlap → midpoint-range → nearest).
   Fixes truncated German translations like "Schal" instead of
   "Schal - die Schals" in session 079cd0d9.

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

klausur-service/backend/cv_words_first.py

@@ -124,13 +124,43 @@ def _cluster_rows(
 # ---------------------------------------------------------------------------
 
 def _assign_word_to_column(word: Dict, columns: List[Dict]) -> int:
-    """Return column index for a word based on its X-center."""
-    x_center = word['left'] + word['width'] / 2
+    """Return column index for a word based on overlap, then center, then nearest.
+
+    Three-pass strategy (consistent with _assign_row_words_to_columns):
+    1. Overlap-based: assign to column with maximum horizontal overlap.
+    2. Midpoint-range: if no overlap, use midpoints between adjacent columns.
+    3. Nearest center: last resort fallback.
+    """
+    w_left = word['left']
+    w_right = w_left + word['width']
+    w_center = w_left + word['width'] / 2
+
+    # Pass 1: overlap-based
+    best_col = -1
+    best_overlap = 0
     for col in columns:
-        if col['x_min'] <= x_center < col['x_max']:
+        overlap = max(0, min(w_right, col['x_max']) - max(w_left, col['x_min']))
+        if overlap > best_overlap:
+            best_overlap = overlap
+            best_col = col['index']
+    if best_col >= 0 and best_overlap > 0:
+        return best_col
+
+    # Pass 2: midpoint-range (non-overlapping assignment zones)
+    for ci, col in enumerate(columns):
+        if ci == 0:
+            assign_left = 0
+        else:
+            assign_left = (columns[ci - 1]['x_max'] + col['x_min']) / 2
+        if ci == len(columns) - 1:
+            assign_right = float('inf')
+        else:
+            assign_right = (col['x_max'] + columns[ci + 1]['x_min']) / 2
+        if assign_left <= w_center < assign_right:
             return col['index']
-    # Fallback: nearest column
-    return min(columns, key=lambda c: abs((c['x_min'] + c['x_max']) / 2 - x_center))['index']
+
+    # Pass 3: nearest column center
+    return min(columns, key=lambda c: abs((c['x_min'] + c['x_max']) / 2 - w_center))['index']
 
 
 def _assign_word_to_row(word: Dict, rows: List[Dict]) -> int: