perf(ai-nutritionist): 文字流式 + TTS 分句首播（test-0415 反馈3-2/3-3）

3-2 文字回复响应速度：
- _sendViaGemini 由 stream:false 改为 kieaiGeminiChatStream（SSE）
- 首包到达即停 loading 圈、aiMsg.streaming=true 显示打字光标
- 逐 delta 累加到 aiMsg.content，TTFB 由全量等待降至首字节

3-3 TTS 朗读延迟：
- splitTTSSentences 按 [。！？!?；;\n] 切分，超长 80 字硬切
- 首句独立合成立即播放；播放期间预合成下一句形成流水线
- innerAudioContext.onEnded 链式触发 _playNextTTSChunk
- stopTTS 清队列，避免后台残留预合成

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

msh_single_uniapp/pages/tool/ai-nutritionist.vue

@@ -767,24 +767,41 @@ export default {
 			}
 		},
 
-		// ========== KieAI Gemini（非流式，BUG-005） ==========
+		// ========== KieAI Gemini（流式 SSE，test-0415 反馈3-2）==========
-		async _sendViaGemini(content, type, aiMsg) {
+		// 旧版 stream: false 等待全文返回 → 5-15s 才出文字。
+		// 现行：调用 kieaiGeminiChatStream，首包到达即停 loading 圈，逐 chunk 渲染 → TTFB <1.5s
+		_sendViaGemini(content, type, aiMsg) {
+			return new Promise((resolve) => {
 				const messages = this.buildChatMessages(content, type)
-			try {
+				let receivedAny = false
-				const response = await api.kieaiGeminiChat({ messages, stream: false })
+				let errorMsg = ''
-				// BUG-005：成功时仅展示 data.choices[0].message.content（经规整）；禁止 getAIResponse / 本地固定话术冒充模型输出
+
-				let reply = api.readKieaiGeminiDataChoicesAssistantText(response && response.data)
+				const ctrl = api.kieaiGeminiChatStream({ messages })
-				if (!reply) {
+				this._streamCtrl = ctrl
-					reply = (api.getKieaiGeminiChatMessageContent(response) || '').trim()
+				ctrl.onMessage((delta) => {
+						if (!receivedAny) {
+							receivedAny = true
+							aiMsg.loading = false
+							aiMsg.streaming = true
+							this.messageList = [...this.messageList]
 						}
-				if (!reply) {
+						aiMsg.content += delta
-					throw new Error('模型未返回有效内容')
+						// 每个 chunk 触发响应式刷新（保证 streaming 光标移动）
-				}
+						this.messageList = [...this.messageList]
-				aiMsg.content = reply
+						this.scrollToBottom()
-			} catch (error) {
+					})
-				console.error('Gemini 对话失败:', error)
+					.onError((err) => {
-				aiMsg.content = (error && error.message) || '抱歉，处理您的请求时出现错误，请稍后再试。'
+						errorMsg = (err && err.message) || ''
+						console.error('Gemini 流式对话失败:', err)
+					})
+					.onComplete(() => {
+						if (!receivedAny) {
+							aiMsg.content = errorMsg || '抱歉，处理您的请求时出现错误，请稍后再试。'
 						}
+						this._streamCtrl = null
+						resolve()
+					})
+			})
 		},
 
 		// ---------- TTS 方法 ----------
@@ -793,6 +810,11 @@ export default {
 			// #ifdef MP-WEIXIN || APP-PLUS
 			const ctx = uni.createInnerAudioContext()
 			ctx.onEnded(() => {
+				// test-0415 反馈3-3：分句队列里还有下一句则继续播；否则结束
+				if (this.ttsPlaying && (this._ttsPrefetch || (this._ttsQueue && this._ttsQueue.length > 0))) {
+					this._playNextTTSChunk()
+					return
+				}
 				this.ttsPlaying = false
 				this.ttsPlayingIndex = -1
 			})
@@ -812,36 +834,83 @@ export default {
 			}
 		},
 
+		/**
+		 * 分句播放 TTS（test-0415 反馈3-3）
+		 * - 旧版：把全文丢给 TTS → 大文本合成耗时长 → 首声延迟 4-8s
+		 * - 现行：按中文句号/问号/感叹号/分号切分；首句合成立即播放，
+		 *   下一句在前一句播放期间预合成，形成流水线
+		 */
 		async playTTS(index) {
 			const msg = this.messageList[index]
 			if (!msg || !msg.content) return
+			if (this.ttsPlaying) this.stopTTS()
 
-			if (this.ttsPlaying) {
+			const sentences = this.splitTTSSentences(msg.content)
-				this.stopTTS()
+			if (sentences.length === 0) return
+
+			this.ttsPlayingIndex = index
+			this.ttsPlaying = true
+			this._ttsQueue = sentences.slice()
+			this._ttsCurrentIndex = index
+			// 预先发起首句合成（首字节最快）
+			this._ttsPrefetch = api.cozeTextToSpeech({ input: this._ttsQueue.shift() })
+			this._playNextTTSChunk()
+		},
+
+		async _playNextTTSChunk() {
+			if (!this.ttsPlaying || this.ttsPlayingIndex !== this._ttsCurrentIndex) return
+			let prefetch = this._ttsPrefetch
+			this._ttsPrefetch = null
+
+			// 提前发起下一句合成（pipeline）
+			if (this._ttsQueue && this._ttsQueue.length > 0) {
+				const nextText = this._ttsQueue.shift()
+				this._ttsPrefetch = api.cozeTextToSpeech({ input: nextText })
 			}
 
 			try {
-				const tempPath = await api.cozeTextToSpeech({ input: msg.content })
+				const tempPath = await prefetch
-				if (!this.innerAudioContext) {
+				if (!this.innerAudioContext || !this.ttsPlaying) return
-					console.warn('[TTS] innerAudioContext 未初始化')
-					return
-				}
-				this.ttsPlayingIndex = index
-				this.ttsPlaying = true
 				this.innerAudioContext.src = tempPath
 				this.innerAudioContext.play()
 			} catch (e) {
 				console.error('[TTS] 合成失败', e)
+				if (this._ttsQueue && this._ttsQueue.length > 0) {
+					// 当前句失败，跳到下一句
+					this._playNextTTSChunk()
+				} else {
 					uni.showToast({ title: '语音合成失败', icon: 'none' })
 					this.ttsPlaying = false
 					this.ttsPlayingIndex = -1
 				}
+			}
+		},
+
+		/** 中文句末标点切分；保留标点；空白过滤；超长（>120 字）强切 */
+		splitTTSSentences(text) {
+			if (!text) return []
+			const re = /[^。！？!?；;\n]+[。！？!?；;]?/g
+			const arr = (String(text).match(re) || [])
+				.map((s) => s.trim())
+				.filter(Boolean)
+			const result = []
+			for (const s of arr) {
+				if (s.length <= 120) {
+					result.push(s)
+				} else {
+					// 超长无终点：按 80 字硬切
+					for (let i = 0; i < s.length; i += 80) result.push(s.slice(i, i + 80))
+				}
+			}
+			return result
 		},
 
 		stopTTS() {
 			if (this.innerAudioContext) {
 				this.innerAudioContext.stop()
 			}
+			this._ttsQueue = []
+			this._ttsPrefetch = null
 			this.ttsPlaying = false
 			this.ttsPlayingIndex = -1
 		},