feat: 语音流式输入管线 + VAD前端集成 + 插件-工具合并清理

- 前端: VAD语音检测(@ricky0123/vad-web) + useVoiceInput双模式(流式WS/REST)
- Gateway: VoiceStreamManager代理WS流式STT到voice-service
- Voice-service: DashScope REST → Realtime WS → Whisper三级引擎 + ffmpeg转码
- 共享模块: pkg/audio(音频转换) + pkg/dashscope(ASR REST客户端)
- 清理: 移除旧plugin-manager和pkg/plugins，完成插件→工具合并
- 文档: 完善gateway-api.md和voice-service.md语音API文档
- 工具: scripts/voice/ 语音转换脚本集

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

backend/gateway/internal/handler/chat_handler.go

@@ -38,6 +38,7 @@ type ChatHandler struct {
 	hub          *ws.Hub
 	sessionStore *store.SessionStore
 	fileStore    *store.FileStore
+	voiceStream  *VoiceStreamManager
 	upgrader     websocket.Upgrader
 	pending      map[string][]queuedMsg // per-session message queue
 	pendingMu    sync.Mutex
@@ -50,6 +51,7 @@ func NewChatHandler(cfg *config.Config, hub *ws.Hub, sessionStore *store.Session
 		hub:          hub,
 		sessionStore: sessionStore,
 		fileStore:    fileStore,
+		voiceStream:  NewVoiceStreamManager(cfg.VoiceServiceURL),
 		pending:      make(map[string][]queuedMsg),
 		upgrader: websocket.Upgrader{
 			ReadBufferSize:  1024,
@@ -131,6 +133,12 @@ func (h *ChatHandler) handleMessage(client *ws.Client, msg ws.ClientMessage) {
 		h.handleChatMessage(client, msg)
 	case "voice_input":
 		h.handleVoiceInput(client, msg)
+	case "voice_stream_start":
+		h.handleVoiceStreamStart(client, msg)
+	case "voice_stream_chunk":
+		h.handleVoiceStreamChunk(client, msg)
+	case "voice_stream_end":
+		h.handleVoiceStreamEnd(client, msg)
 	case "history":
 		h.handleHistoryRequest(client, msg)
 	default:
@@ -436,11 +444,13 @@ func (h *ChatHandler) streamResponse(client *ws.Client, mode string, reqBody []b
 		// 处理审查后的结构化消息 (review)
 		if len(chunk.ReviewMessages) > 0 {
 			for i, rm := range chunk.ReviewMessages {
+				msgType := rm.Type
+				if msgType == "" {
+					msgType = "chat"
+				}
 				role := "assistant"
-				msgType := "chat"
-				if rm.Type == "action" {
+				if msgType == "action" {
 					role = "action"
-					msgType = "action"
 				}
 				reviewMsgID := fmt.Sprintf("%s_r%d", msgID, i)
 				// 持久化每条审查消息 (action 角色映射为 assistant，LLM 模型不支持自定义角色)
@@ -473,6 +483,7 @@ func (h *ChatHandler) streamResponse(client *ws.Client, mode string, reqBody []b
 					SessionID:  client.SessionID,
 					Timestamp:  time.Now().UnixMilli(),
 					ClientInfo: clientInfo,
+					Metadata:   rm.Metadata,
 				})
 				// 使用 MessageScheduler 计算的 per-message 延迟
 				if rm.DelayMs > 0 {
@@ -650,6 +661,96 @@ func (h *ChatHandler) handleVoiceInput(client *ws.Client, msg ws.ClientMessage)
 	}()
 }
 
+// handleVoiceStreamStart begins a streaming voice session via voice-service.
+func (h *ChatHandler) handleVoiceStreamStart(client *ws.Client, msg ws.ClientMessage) {
+	format := msg.Format
+	if format == "" {
+		format = "webm"
+	}
+	language := msg.Language
+	if language == "" {
+		language = "zh"
+	}
+
+	if err := h.voiceStream.StartStream(client, format, language); err != nil {
+		logger.Printf("[voice-stream] 启动流式 STT 失败: %v", err)
+		client.SendMessage(ws.ServerMessage{
+			Type:      "error",
+			MessageID: "msg_" + generateID(),
+			Error:     "启动语音流失败: " + err.Error(),
+			Timestamp: time.Now().UnixMilli(),
+		})
+		return
+	}
+
+	client.SendMessage(ws.ServerMessage{
+		Type:      "voice_interim",
+		MessageID: "voice_" + generateID(),
+		Text:      "",
+		Timestamp: time.Now().UnixMilli(),
+	})
+}
+
+// handleVoiceStreamChunk forwards an audio chunk to the active voice stream.
+func (h *ChatHandler) handleVoiceStreamChunk(client *ws.Client, msg ws.ClientMessage) {
+	if msg.AudioData == "" {
+		return
+	}
+
+	audioData, err := decodeBase64(msg.AudioData)
+	if err != nil {
+		logger.Printf("[voice-stream] 解码音频块失败: %v", err)
+		return
+	}
+
+	if err := h.voiceStream.SendChunk(client.ClientID, client.SessionID, audioData, msg.Sequence); err != nil {
+		logger.Printf("[voice-stream] 发送音频块失败: %v", err)
+	}
+}
+
+// handleVoiceStreamEnd stops the voice stream and processes the final transcription.
+func (h *ChatHandler) handleVoiceStreamEnd(client *ws.Client, msg ws.ClientMessage) {
+	go func() {
+		text, err := h.voiceStream.EndStream(client.ClientID, client.SessionID)
+		if err != nil {
+			logger.Printf("[voice-stream] 结束流式 STT 失败: %v", err)
+			client.SendMessage(ws.ServerMessage{
+				Type:      "error",
+				MessageID: "msg_" + generateID(),
+				Error:     "语音流处理失败: " + err.Error(),
+				Timestamp: time.Now().UnixMilli(),
+			})
+			return
+		}
+
+		if text == "" {
+			client.SendMessage(ws.ServerMessage{
+				Type:      "voice_final",
+				MessageID: "voice_" + generateID(),
+				Text:      "",
+				Timestamp: time.Now().UnixMilli(),
+			})
+			return
+		}
+
+		// Send final transcription to frontend
+		client.SendMessage(ws.ServerMessage{
+			Type:      "voice_final",
+			MessageID: "voice_" + generateID(),
+			Text:      text,
+			Timestamp: time.Now().UnixMilli(),
+		})
+
+		// Route the transcribed text as a regular chat message to ai-core
+		chatMsg := ws.ClientMessage{
+			Type:    "message",
+			Content: text,
+			Mode:    msg.Mode,
+		}
+		h.handleChatMessage(client, chatMsg)
+	}()
+}
+
 // transcribeAudio 将 base64 编码的音频发送到 voice-service 进行转录。
 func (h *ChatHandler) transcribeAudio(audioB64 string, format string) (string, error) {
 	audioData, err := decodeBase64(audioB64)