fix(it0_app): stop using systemPrompt as conversation title

Voice sessions set systemPrompt to the voice-mode instruction string,
causing every voice conversation to display '你正在通过语音与用户实时对话。请…'
as its title in the chat history list.

Title derivation priority (highest to lowest):
  1. metadata.title  — explicit title saved by backend on first task
  2. metadata.voiceMode == true → '语音对话 M/D HH:mm'
  3. Fallback → '对话 M/D HH:mm' based on session createdAt

it0_app/lib/features/chat/presentation/widgets/conversation_drawer.dart

@@ -22,7 +22,11 @@ class SessionSummary {
   factory SessionSummary.fromJson(Map<String, dynamic> json) {
     return SessionSummary(
       id: json['id'] as String,
-      title: json['systemPrompt'] as String? ?? _generateTitle(json),
+      // NOTE: Never use systemPrompt as title — it's an internal engine instruction
+      // (voice sessions set systemPrompt to the voice-mode rules string, which caused
+      // every voice conversation to appear as "你正在通过语音与用户实时对话。请…").
+      // Always derive the display title from metadata or the session creation time.
+      title: _generateTitle(json),
       status: json['status'] as String? ?? 'active',
       createdAt: DateTime.tryParse(json['createdAt'] as String? ?? '') ?? DateTime.now(),
       updatedAt: DateTime.tryParse(json['updatedAt'] as String? ?? '') ?? DateTime.now(),
@@ -30,12 +34,21 @@ class SessionSummary {
   }
 
   static String _generateTitle(Map<String, dynamic> json) {
-    // Use metadata or fallback to date-based title
     final meta = json['metadata'] as Map<String, dynamic>?;
-    if (meta != null && meta['title'] != null) {
+
-      return meta['title'] as String;
+    // 1. Use the explicit title stored by the backend on first task (text sessions)
-    }
+    final storedTitle = meta?['title'] as String?;
+    if (storedTitle != null && storedTitle.isNotEmpty) return storedTitle;
+
     final createdAt = DateTime.tryParse(json['createdAt'] as String? ?? '');
+
+    // 2. Voice sessions: prefix with 🎙 and show time
+    final isVoice = meta?['voiceMode'] as bool? ?? false;
+    if (isVoice && createdAt != null) {
+      return '语音对话 ${createdAt.month}/${createdAt.day} ${createdAt.hour}:${createdAt.minute.toString().padLeft(2, '0')}';
+    }
+
+    // 3. Fallback: date-based title
     if (createdAt != null) {
       return '对话 ${createdAt.month}/${createdAt.day} ${createdAt.hour}:${createdAt.minute.toString().padLeft(2, '0')}';
     }