refactor: 优化创建回忆流程

- 优化CreateMemoryScreen创建回忆页面
- 优化CreateMemoryViewModel创建回忆ViewModel

Co-authored-by: Cursor <cursoragent@cursor.com>

app-android/app/src/main/java/com/huaga/life_echo/ui/screens/CreateMemoryScreen.kt

@@ -29,6 +29,7 @@ import com.huaga.life_echo.network.models.MessageDto
 import com.huaga.life_echo.ui.components.chat.ChatHeader
 import com.huaga.life_echo.ui.components.chat.ChatInputField
 import com.huaga.life_echo.ui.components.chat.MessageList
+import com.huaga.life_echo.ui.components.chat.ReleaseAction
 import com.huaga.life_echo.ui.components.chat.RecordingOverlayContent
 import com.huaga.life_echo.ui.components.chat.RecordingState
 import com.huaga.life_echo.ui.components.debug.WebSocketDebugPanel
@@ -159,7 +160,8 @@ fun CreateMemoryScreen(
         ChatHeader(
             title = "回忆录助手",
             isOnline = connectionStatus == "已连接",
-            onBackClick = { navController?.popBackStack() }
+            onBackClick = { navController?.popBackStack() },
+            onNewConversationClick = { viewModel.startConversation() }
         )
         
         // WebSocket调试面板（仅在开发模式下显示）
@@ -208,7 +210,8 @@ fun CreateMemoryScreen(
                     keyboardController?.hide()
                 }
             },
-            enabled = !isStreaming && !isVoiceRecording,
+            // 录音过程中也必须保持语音按钮可用，否则会丢失手指抬起事件，导致无法停止录音
+            enabled = !isStreaming,
             // 语音相关回调
             onStartRecording = {
                 if (hasRecordPermission && Build.VERSION.SDK_INT >= Build.VERSION_CODES.O) {
@@ -226,11 +229,17 @@ fun CreateMemoryScreen(
             recordingDuration = recordingDuration
         )
     }
-        // 录音浮层：仅视觉反馈，不消费触摸，松开由按钮接收并立即发送
+        // 录音浮层：覆盖在聊天之上，只检测本层手势；左滑取消、右滑转文字、否则松开发送语音
         if (recordingOverlayState != RecordingState.IDLE) {
             RecordingOverlayContent(
-                state = recordingOverlayState,
                 duration = recordingDuration,
+                onRelease = { action ->
+                    when (action) {
+                        ReleaseAction.CANCEL -> viewModel.cancelRecordingVoice()
+                        ReleaseAction.SEND_VOICE -> viewModel.stopAndSendRecording()
+                        ReleaseAction.SEND_AS_TEXT -> viewModel.stopAndSendRecordingAsText()
+                    }
+                },
                 modifier = Modifier.fillMaxSize()
             )
         }

app-android/app/src/main/java/com/huaga/life_echo/ui/viewmodel/CreateMemoryViewModel.kt

@@ -19,11 +19,13 @@ import com.huaga.life_echo.network.ApiService
 import com.huaga.life_echo.network.AuthService
 import com.huaga.life_echo.network.models.MessageDto
 import com.huaga.life_echo.data.database.Chapter
+import kotlinx.coroutines.channels.Channel
 import kotlinx.coroutines.flow.MutableStateFlow
 import kotlinx.coroutines.flow.StateFlow
 import kotlinx.coroutines.flow.asStateFlow
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.delay
+import kotlinx.coroutines.withTimeoutOrNull
 
 class CreateMemoryViewModel(
     private val conversationRepository: ConversationRepository,
@@ -97,6 +99,11 @@ class CreateMemoryViewModel(
     private val _audioDurations = MutableStateFlow<Map<String, Int>>(emptyMap())
     val audioDurations: StateFlow<Map<String, Int>> = _audioDurations.asStateFlow()
     
+    // 仅转写结果：发 transcribe_only 后等待 transcript 用
+    private val pendingTranscribeChannel = Channel<String>(Channel.RENDEZVOUS)
+    @Volatile
+    private var waitingForTranscribeOnly = false
+    
     init {
         TokenManager.initialize(context)
     }
@@ -182,6 +189,16 @@ class CreateMemoryViewModel(
                 isRecording.value = true
                 connectionStatus.value = "连接中..."
                 
+                // 新建对话：清空当前会话展示状态
+                historyMessages.value = emptyList()
+                userMessages.value = emptyList()
+                streamingText.value = ""
+                agentResponse.value = ""
+                isStreaming.value = false
+                isTyping.value = false
+                _audioFilePaths.value = emptyMap()
+                _audioDurations.value = emptyMap()
+                
                 // 清除旧的任务记录
                 apiService.clearTasks()
                 
@@ -244,24 +261,20 @@ class CreateMemoryViewModel(
     }
     
     /**
-     * 停止录音并发送
+     * 停止录音并发送（手势释放后立即调用，内部在协程中执行避免阻塞触摸回调，马上发送对应语音）
      */
     fun stopAndSendRecording() {
-        Log.d(TAG, "停止录音并发送")
-        val result = voiceRecorder.stopRecording()
-        
-        if (result == null) {
-            Log.e(TAG, "录音失败，result 为空")
-            return
-        }
-        
-        if (result.durationSeconds < MIN_RECORDING_DURATION) {
-            Log.w(TAG, "录音时间太短: ${result.durationSeconds}s")
-            // 可以在这里添加提示用户的逻辑
-            return
-        }
-        
         viewModelScope.launch {
+            Log.d(TAG, "停止录音并发送")
+            val result = voiceRecorder.stopRecording()
+            if (result == null) {
+                Log.e(TAG, "录音失败，result 为空")
+                return@launch
+            }
+            if (result.durationSeconds < MIN_RECORDING_DURATION) {
+                Log.w(TAG, "录音时间太短: ${result.durationSeconds}s")
+                return@launch
+            }
             sendAudioMessage(result.audioBytes, result.filePath, result.durationSeconds)
         }
     }
@@ -274,17 +287,75 @@ class CreateMemoryViewModel(
         voiceRecorder.cancelRecording()
     }
     
+    /**
+     * 停止录音并转为文字发送（用于浮层「转文字」）
+     */
+    fun stopAndSendRecordingAsText() {
+        viewModelScope.launch {
+            Log.d(TAG, "停止录音并转文字发送")
+            val result = voiceRecorder.stopRecording() ?: run {
+                Log.e(TAG, "录音结果为空")
+                return@launch
+            }
+            if (result.durationSeconds < MIN_RECORDING_DURATION) {
+                Log.w(TAG, "录音太短: ${result.durationSeconds}s")
+                return@launch
+            }
+            if (conversationId.value == null) {
+                startConversation()
+                delay(300)
+            }
+            if (!webSocketClient.isConnected()) {
+                connectionStatus.value = "未连接，正在重连..."
+                conversationId.value?.let { id ->
+                    val token = TokenManager.getAccessToken()
+                    try {
+                        webSocketClient.connect(
+                            id,
+                            token,
+                            onMessage = { m -> handleWebSocketMessage(m) },
+                            onError = { connectionStatus.value = "错误: $it" }
+                        )
+                        delay(300)
+                    } catch (e: Exception) {
+                        connectionStatus.value = "重连失败: ${e.message}"
+                        return@launch
+                    }
+                } ?: run {
+                    connectionStatus.value = "对话ID为空"
+                    return@launch
+                }
+            }
+            val id = conversationId.value ?: return@launch
+            try {
+                waitingForTranscribeOnly = true
+                webSocketClient.sendTranscribeOnly(result.audioBytes, id)
+                val text = withTimeoutOrNull(15000L) { pendingTranscribeChannel.receive() }
+                waitingForTranscribeOnly = false
+                if (!text.isNullOrBlank() && !text.startsWith("转写失败")) {
+                    sendTextMessage(text)
+                } else {
+                    connectionStatus.value = "转写失败或超时，请重试"
+                }
+            } catch (e: Exception) {
+                waitingForTranscribeOnly = false
+                Log.e(TAG, "转文字失败: ${e.message}", e)
+                connectionStatus.value = "转写失败: ${e.message}"
+            }
+        }
+    }
+    
     /**
      * 发送音频消息
      */
     private suspend fun sendAudioMessage(audioBytes: ByteArray, filePath: String, durationSeconds: Int) {
         Log.d(TAG, "准备发送音频消息，大小: ${audioBytes.size}, 时长: ${durationSeconds}s")
         
-        // 确保已连接
+        // 确保已连接（手势松开后马上发语音，尽量缩短等待）
         if (conversationId.value == null) {
             Log.d(TAG, "对话ID为空，开始创建新对话")
             startConversation()
-            delay(500)
+            delay(300)
         }
         
         // 检查连接状态
@@ -301,7 +372,7 @@ class CreateMemoryViewModel(
                         onMessage = { message -> handleWebSocketMessage(message) },
                         onError = { errorMsg -> connectionStatus.value = "错误: $errorMsg" }
                     )
-                    delay(500)
+                    delay(300)
                 } catch (e: Exception) {
                     connectionStatus.value = "重连失败: ${e.message}"
                     return
@@ -311,7 +382,7 @@ class CreateMemoryViewModel(
                 return
             }
         }
-        
+
         conversationId.value?.let { id ->
             // 生成临时消息 ID
             val tempMessageId = "audio_user_${System.currentTimeMillis()}"
@@ -461,10 +532,13 @@ class CreateMemoryViewModel(
         
         when (message.type) {
             MessageType.transcript -> {
-                // 语音转文字结果
                 val text = message.getString("text") ?: ""
                 transcript.value = text
                 Log.d(TAG, "收到语音转文字结果: $text")
+                if (waitingForTranscribeOnly) {
+                    waitingForTranscribeOnly = false
+                    pendingTranscribeChannel.trySend(text)
+                }
             }
             MessageType.agent_response -> {
                 // 处理Agent回复（可能有多条消息，每条作为单独气泡显示）