fix: Zeilen an Box-Grenze nicht mehr abschneiden (border_thickness Margin)

- detect_rows: Content-Strips nutzen jetzt box_ranges_inner (geschrumpft
  um border_thickness, min 5px) statt der vollen Box-Range
- detect_words: _row_in_box Filter nutzt ebenfalls inner Range
- Dadurch wird die letzte Zeile oberhalb einer Box nicht mehr
  faelschlicherweise der Box zugeordnet und ausgeschlossen

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

klausur-service/backend/ocr_pipeline_api.py

@@ -1627,19 +1627,28 @@ async def detect_rows(session_id: str):
     else:
         zones = column_result.get("zones") or []  # zones can be None for sub-sessions
 
-        # Collect box y-ranges for filtering
+        # Collect box y-ranges for filtering.
+        # Use border_thickness to shrink the exclusion zone: the border pixels
+        # belong visually to the box frame, but text rows above/below the box
+        # may overlap with the border area and must not be clipped.
         box_ranges = []  # [(y_start, y_end)]
+        box_ranges_inner = []  # [(y_start + border, y_end - border)] for row filtering
         for zone in zones:
             if zone.get("zone_type") == "box" and zone.get("box"):
                 box = zone["box"]
+                bt = max(box.get("border_thickness", 0), 5)  # minimum 5px margin
                 box_ranges.append((box["y"], box["y"] + box["height"]))
+                # Inner range: shrink by border thickness so boundary rows aren't excluded
+                box_ranges_inner.append((box["y"] + bt, box["y"] + box["height"] - bt))
 
         if box_ranges and inv is not None:
             # Combined-image approach: strip box regions from inv image,
             # run row detection on the combined image, then remap y-coords back.
             content_strips = []  # [(y_start, y_end)] in absolute coords
-            # Build content strips by subtracting box ranges from [top_y, bottom_y]
-            sorted_boxes = sorted(box_ranges, key=lambda r: r[0])
+            # Build content strips by subtracting box inner ranges from [top_y, bottom_y].
+            # Using inner ranges means the border area is included in the content
+            # strips, so the last row above a box isn't clipped by the border.
+            sorted_boxes = sorted(box_ranges_inner, key=lambda r: r[0])
             strip_start = top_y
             for by_start, by_end in sorted_boxes:
                 if by_start > strip_start:
@@ -1934,18 +1943,21 @@ async def detect_words(
             ]
             row.word_count = len(row.words)
 
-    # Exclude rows that fall within box zones
+    # Exclude rows that fall within box zones.
+    # Use inner box range (shrunk by border_thickness) so that rows at
+    # the boundary (overlapping with the box border) are NOT excluded.
     zones = column_result.get("zones") or []
-    box_ranges = []
+    box_ranges_inner = []
     for zone in zones:
         if zone.get("zone_type") == "box" and zone.get("box"):
             box = zone["box"]
-            box_ranges.append((box["y"], box["y"] + box["height"]))
+            bt = max(box.get("border_thickness", 0), 5)  # minimum 5px margin
+            box_ranges_inner.append((box["y"] + bt, box["y"] + box["height"] - bt))
 
-    if box_ranges:
+    if box_ranges_inner:
         def _row_in_box(r):
             center_y = r.y + r.height / 2
-            return any(by_s <= center_y < by_e for by_s, by_e in box_ranges)
+            return any(by_s <= center_y < by_e for by_s, by_e in box_ranges_inner)
 
         before_count = len(row_geoms)
         row_geoms = [r for r in row_geoms if not _row_in_box(r)]