feat: capture reasoning content from thinking models, improve thinking UI

- Backend: intercept reasoning_content from DeepSeek-R1/QwQ streaming
  chunks and emit as thinking events (stage=reasoning)
- RAG chain: show retrieved document titles/previews in thinking steps
- Conversation chain: directly stream from LLM to capture reasoning
- Frontend: collapsible thinking panel with reasoning section,
  document details, and time summary
- Replace relative time with HH:mm format for message timestamps

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/src/rag/conversation_chains.py

@@ -80,24 +80,41 @@ class ConversationChain:
         }
     
     async def astream_with_thinking(self, question: str, chat_history: List[Dict] = None):
-        """流式调用（包含思考过程）"""
+        """流式调用（包含思考过程，捕获推理模型的真实推理内容）"""
         import time
-        
+
         # 思考阶段
         start_time = time.time()
         yield {"type": "thinking", "stage": "understanding", "message": "正在理解问题..."}
-        
+
         # 准备历史
         history_messages = self._format_history(chat_history or [])
-        
+        history_count = len([m for m in (chat_history or []) if m["role"] == "user"])
+        yield {"type": "thinking", "stage": "preparing", "message": f"加载对话上下文（{history_count} 轮历史）..." if history_count > 0 else "准备生成回答..."}
+
         yield {"type": "thinking", "stage": "generating", "message": "正在生成回答..."}
-        
-        # 流式生成
-        async for chunk in self.chain.astream({
+
+        # 直接流式调用 LLM 以捕获推理内容
+        prompt_messages = await self.prompt.ainvoke({
             "question": question,
             "chat_history": history_messages
-        }):
-            yield {"type": "chunk", "content": chunk}
+        })
+
+        content_started = False
+        async for chunk in self.llm.astream(prompt_messages):
+            # 捕获推理内容（DeepSeek-R1/QwQ 等推理模型）
+            if hasattr(chunk, 'additional_kwargs') and 'reasoning_content' in chunk.additional_kwargs:
+                reasoning_text = chunk.additional_kwargs['reasoning_content']
+                if reasoning_text:
+                    yield {"type": "thinking", "stage": "reasoning", "message": reasoning_text}
+            elif hasattr(chunk, 'reasoning_content') and chunk.reasoning_content:
+                yield {"type": "thinking", "stage": "reasoning", "message": chunk.reasoning_content}
+
+            content = chunk.content if hasattr(chunk, 'content') else str(chunk)
+            if content:
+                if not content_started:
+                    content_started = True
+                yield {"type": "chunk", "content": content}
         
         # 完成
         total_time = time.time() - start_time