Fix: Separate thinking and summary fields during streaming

Previously:
- Thinking field accumulated ALL tokens (thinking + summary)
- Summary field showed parsed summary
- Result: Both fields showed overlapping content

Now:
- Thinking field shows ONLY thinking blocks (content inside tags)
- Summary field shows ONLY summary (content outside tags)
- Both fields update in real-time during streaming

Changes:
- Remove token accumulation in thinking field (line 162)
- Parse and update both fields on every token
- Always update both fields (not just when truthy)
- Ensures clean separation of thinking vs summary content

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

app.py

@@ -157,25 +157,24 @@ def summarize_streaming(file_obj, max_tokens: int = 2048, temperature: float = 0
                     # Convert to Traditional Chinese (Taiwan)
                     converted = converter.convert(content)
                     full_response += converted
-                    
-                    # Show raw output in thinking field (token by token)
-                    current_thinking += converted
-                    
-                    # Try to extract summary (content outside thinking tags)
+
+                    # Parse thinking blocks and summary from accumulated response
                     thinking_blocks, summary = parse_thinking_blocks(full_response)
-                    if summary:
-                        current_summary = warning_msg + summary
-                    
+
+                    # Update thinking field (only show thinking blocks, not raw stream)
+                    current_thinking = thinking_blocks if thinking_blocks else ""
+
+                    # Update summary field (only show summary, not thinking blocks)
+                    current_summary = warning_msg + summary if summary else warning_msg
+
                     # Yield both fields on every token
                     yield (current_thinking, current_summary)
         
-        # Final parse at the end
+        # Final parse to ensure consistency (redundant but safe)
         final_thinking, final_summary = parse_thinking_blocks(full_response)
-        if final_thinking:
-            current_thinking = final_thinking
-        if final_summary:
-            current_summary = warning_msg + final_summary
-        
+        current_thinking = final_thinking if final_thinking else ""
+        current_summary = warning_msg + final_summary if final_summary else warning_msg
+
         # Final yield
         yield (current_thinking, current_summary)