fix: relax graphic detection for small circles/balloons

- Lower min_area from 200 to 80 (small balloons ~100-300px²)
- Lower word_pad from 10 to 5 (10px was eating nearby graphics)
- Relax circle detection: circularity>0.55, min_dim>15 (was 0.70/25)
- Text fragments still filtered by _classify_shape noise threshold
- Add ACCEPT logging for debugging

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

klausur-service/backend/cv_graphic_detect.py

@@ -122,8 +122,8 @@ def _classify_shape(
     max_dim = max(bw, bh)
 
     # --- Circle / balloon --- (check first, most reliable)
-    # Must be reasonably large (not a dot/period)
-    if circularity > 0.70 and 0.6 < aspect < 1.7 and min_dim > 25:
+    # Must be reasonably large (not a dot/period) — min 15px
+    if circularity > 0.55 and 0.5 < aspect < 2.0 and min_dim > 15:
         conf = min(0.95, circularity)
         return "circle", conf
 
@@ -169,9 +169,9 @@ def detect_graphic_elements(
     img_bgr: np.ndarray,
     word_boxes: List[Dict],
     detected_boxes: Optional[List[Dict]] = None,
-    min_area: int = 200,
+    min_area: int = 80,
     max_area_ratio: float = 0.25,
-    word_pad: int = 10,
+    word_pad: int = 5,
     max_elements: int = 50,
 ) -> List[GraphicElement]:
     """Find non-text graphical elements on the page.
@@ -184,9 +184,9 @@ def detect_graphic_elements(
         img_bgr: BGR color image.
         word_boxes: List of OCR word dicts with left/top/width/height.
         detected_boxes: Optional list of detected box dicts (x/y/w/h).
-        min_area: Minimum contour area to keep (200 filters text fragments).
+        min_area: Minimum contour area to keep (80 filters tiny noise).
         max_area_ratio: Maximum area as fraction of image area.
-        word_pad: Padding around word boxes for exclusion (10px covers font edges).
+        word_pad: Padding around word boxes for exclusion (5px).
         max_elements: Maximum number of elements to return.
 
     Returns:
@@ -306,6 +306,8 @@ def detect_graphic_elements(
         masked_hsv = roi_hsv[cnt_mask > 0]
         color_name, color_hex = _dominant_color(masked_hsv)
 
+        logger.info("GraphicDetect ACCEPT: %s at (%d,%d) %dx%d area=%d color=%s conf=%.2f",
+                    shape, bx, by, bw, bh, int(area), color_name, conf)
         candidates.append(GraphicElement(
             x=bx, y=by, width=bw, height=bh,
             area=int(area),