fix: extend tiny symbol filter to all non-black colors, raise area to 200

Step 5i rule (a) only caught blue tiny symbols. Graphic fragments from
page illustrations (e.g. orange quote mark from man illustration) were
missed. Now filters any non-black colored word_box with area < 200 and
confidence < 85.

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

klausur-service/backend/grid_editor_api.py

@@ -2292,7 +2292,7 @@ async def _build_grid_core(session_id: str, session: dict) -> dict:
     # OCR reads these as text artifacts (©, e, *, or even plausible words
     # like "fighily" overlapping the real word "tightly").
     # Detection rules:
-    #   a) Tiny blue symbols: area < 150 AND conf < 85
+    #   a) Tiny coloured symbols: area < 200 AND conf < 85 (any non-black)
     #   b) Overlapping word_boxes: >40% x-overlap → remove lower confidence
     #   c) Duplicate text: consecutive blue wbs with identical text, gap < 6px
     bullet_removed = 0
@@ -2303,10 +2303,11 @@ async def _build_grid_core(session_id: str, session: dict) -> dict:
                 continue
             to_remove: set = set()
 
-            # Rule (a): tiny blue symbols
+            # Rule (a): tiny coloured symbols (bullets, graphic fragments)
             for i, wb in enumerate(wbs):
-                if (wb.get("color_name") == "blue"
-                        and wb.get("width", 0) * wb.get("height", 0) < 150
+                cn = wb.get("color_name", "black")
+                if (cn != "black"
+                        and wb.get("width", 0) * wb.get("height", 0) < 200
                         and wb.get("conf", 100) < 85):
                     to_remove.add(i)