fix: qwen thinking toggle and recall log styles

modules/story-summary/vector/llm/llm-service.js

@@ -29,7 +29,7 @@ function b64UrlEncode(str) {
 
 /**
  * 统一LLM调用 - 走酒馆后端（非流式）
- * 修复：assistant prefill 用 bottomassistant 参数传递
+ * assistant prefill 用 bottomassistant 参数传递
  */
 export async function callLLM(messages, options = {}) {
     const {
@@ -46,10 +46,10 @@ export async function callLLM(messages, options = {}) {
         throw new Error('L0 requires siliconflow API key');
     }
 
-    // ★ 关键修复：分离 assistant prefill
+    // 分离 assistant prefill
     let topMessages = [...messages];
     let assistantPrefill = '';
-    
+
     if (topMessages.length > 0 && topMessages[topMessages.length - 1]?.role === 'assistant') {
         const lastMsg = topMessages.pop();
         assistantPrefill = lastMsg.content || '';
@@ -70,6 +70,10 @@ export async function callLLM(messages, options = {}) {
         apipassword: apiKey,
         model: DEFAULT_L0_MODEL,
     };
+    const isQwen3 = String(DEFAULT_L0_MODEL || '').includes('Qwen3');
+    if (isQwen3) {
+        args.enable_thinking = 'false';
+    }
 
     // ★ 用 bottomassistant 参数传递 prefill
     if (assistantPrefill) {