整理多轮对话发送时的数据结构，减少tokend冗余消耗

src/components/AICard.vue

@@ -405,28 +405,6 @@ const mergeReferences = (...groups: Conversation[][]) => {
     return Array.from(uniqueReferences.values());
 };
 
-const buildReferenceContextMessage = (references: Conversation[]) => {
-    if (!references.length) {
-        return null;
-    }
-
-    return {
-        role: 'system' as const,
-        content: [
-            '以下内容是用户通过 @ 主动引用的历史对话上下文。',
-            '这些内容仅作为当前问题的补充背景，不要把它们当作需要继续续写的对话。',
-            '如果当前问题明显依赖这些引用，请优先结合引用内容作答。',
-            '',
-            ...references.flatMap((item, index) => [
-                `[@引用 ${index + 1}]`,
-                `问题：${item.prompt}`,
-                `回答：${item.answer}`,
-                '',
-            ]),
-        ].join('\n'),
-    };
-};
-
 const submitPrompt = async (
     promptText: string,
     references: Conversation[] = selectedReferences.value,
@@ -460,10 +438,8 @@ const submitPrompt = async (
 
         let fullResponse = '';
         const orderedReferences = [...references].sort((left, right) => Number(left.id_timestamp) - Number(right.id_timestamp));
-        const referenceContextMessage = buildReferenceContextMessage(orderedReferences);
         const messages = [
             {role: 'system' as const, content:'你是一个AI助手，请根据用户的问题给出回答'},
-            ...(referenceContextMessage ? [referenceContextMessage] : []),
             ...orderedReferences.flatMap((item) => {
                 return [
                     { role: 'user' as const, content: item.prompt },