feat(core): add ModelScope support for TTS and new office load status

Add support to download and load TTS model from ModelScope, with a fallback to the HuggingFace mirror.
Implement a `documentLoadStatus` property and helper functions in `office.js` to track file loading state.
Improve request cancellation logic in `api.js`, ensuring proper cancel URL resolution and request‑id handling.

These changes enhance robustness, reduce external dependencies, and provide better UX for office file handling.

backend/tts_asr.py

@@ -25,6 +25,10 @@ router = APIRouter()
 # Global TTS model instance
 _tts_model: Optional["Qwen3TTSModel"] = None
 
+# Model paths for loading
+MODEL_ID_HF = "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign"
+MODEL_ID_MS = "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign"
+
 
 def _get_device_map() -> str:
     """设备检测逻辑：优先 CUDA，其次 MPS，最后 CPU"""
@@ -38,6 +42,24 @@ def _get_device_map() -> str:
     return "cpu"
 
 
+def _download_model_from_modelscope() -> Optional[str]:
+    """从 ModelScope 下载模型到本地临时目录"""
+    try:
+        from modelscope import snapshot_download
+        cache_dir = os.path.join(os.path.dirname(__file__), "models")
+        os.makedirs(cache_dir, exist_ok=True)
+        model_dir = snapshot_download(
+            MODEL_ID_MS,
+            cache_dir=cache_dir,
+            revision="master"
+        )
+        logger.info("ModelScope 模型下载完成: %s", model_dir)
+        return model_dir
+    except Exception as e:
+        logger.warning("ModelScope 下载失败: %s", e)
+        return None
+
+
 async def _warmup_tts():
     """预热 TTS 模型"""
     await asyncio.to_thread(_load_tts_model_with_retry)
@@ -58,28 +80,42 @@ def _load_tts_model_with_retry(max_retries: int = 3) -> "Qwen3TTSModel":
     if Qwen3TTSModel is None:
         raise RuntimeError("qwen_tts 库未安装，无法加载 TTS 模型")
 
-    candidates = [
-        "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
-        "ModelScope/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
-    ]
     device_map = _get_device_map()
     last_err = None
-    for i, model_id in enumerate(candidates, start=1):
+
+    # 策略1: 尝试从 ModelScope 下载后加载
+    for attempt in range(max_retries):
         try:
+            logger.info("尝试从 ModelScope 下载模型...")
+            model_path = _download_model_from_modelscope()
+            if model_path and os.path.isdir(model_path):
+                _tts_model = Qwen3TTSModel.from_pretrained(
+                    model_path,
+                    device_map=device_map,
+                    dtype=torch.float16,
+                )
+                logger.info("ModelScope 模型加载成功: %s", model_path)
+                return _tts_model
+        except Exception as e:
+            logger.warning("ModelScope 加载失败 (尝试 %d/%d): %s", attempt + 1, max_retries, e)
+            last_err = e
+
+    # 策略2: 尝试从 HuggingFace 镜像加载
+    for attempt in range(max_retries):
+        try:
+            logger.info("尝试从 HuggingFace 镜像加载模型...")
             _tts_model = Qwen3TTSModel.from_pretrained(
-                model_id,
+                MODEL_ID_HF,
                 device_map=device_map,
                 dtype=torch.float16,
-                attn_implementation="flash_attention_2",
             )
-            logger.info("Loaded TTS model from %s", model_id)
+            logger.info("HuggingFace 模型加载成功")
             return _tts_model
         except Exception as e:
-            logger.warning("Failed to load TTS model from %s: %s", model_id, e)
+            logger.warning("HuggingFace 加载失败 (尝试 %d/%d): %s", attempt + 1, max_retries, e)
             last_err = e
-            if i >= max_retries:
-                break
-    raise RuntimeError(f"Unable to load TTS model from sources: {candidates}") from last_err
+
+    raise RuntimeError(f"无法加载 TTS 模型: {last_err}") from last_err
 
 
 class TTSRequest(BaseModel):
@@ -160,10 +196,10 @@ async def tts_endpoint(req: TTSRequest):
     speaker = req.speaker or "Vivian"
 
     try:
-        wavs_sr = model.generate_custom_voice(
+        # VoiceDesign 模型使用 generate_voice_design 方法
+        wavs_sr = model.generate_voice_design(
             text=text,
             language="Chinese",
-            speaker=speaker,
             instruct=instruct,
         )
     except Exception as e: