feat: 配置写死与 baked 模块，Alembic 建表，百度仅 BAIDU_*

- 新增 app/baked/algorithm|pipeline，非部署参数不再走 env；Settings 保留 DB/HTTP/RTSP/海康/百度/MinIO/Demo
- 移除 init_db_schema 与 reload 配置；main 仅 check_database；start*.sh 在 uvicorn 前执行 alembic upgrade head
- 依赖 psycopg[binary] 供 Alembic 同步 URL；alembic/env 注释与预发清单更新
- 撕段门控消费管线、各视频/语音/归档调用改为 baked
- 百度环境变量仅 BAIDU_APP_ID、BAIDU_API_KEY、BAIDU_SECRET_KEY 与 BAIDU_* 超时/ASR；人脸脚本与 baidu_speech 文案同步
- 全量单测与 .env.example 更新；.gitignore 忽略 refs/（本地权重/视频不入库）

Made-with: Cursor

app/services/consumable_vision_algorithm.py

@@ -19,7 +19,7 @@ import yaml
 from loguru import logger
 from ultralytics import YOLO
 
-from app.config import Settings, settings
+from app.baked import algorithm as ba
 
 
 def _ensure_yolo_config_dir() -> None:
@@ -361,14 +361,20 @@ def window_bucket_to_best_snap(
 class ConsumableVisionAlgorithmService:
     """手部检测（可选）+ 耗材分类；供 CameraSessionManager 在视频线程中调用。"""
 
-    def __init__(self, app_settings: Settings | None = None) -> None:
+    def __init__(self, *, labels_yaml_path: str | None = None) -> None:
         _ensure_yolo_config_dir()
-        self._s = app_settings or settings
+        self._labels_yaml_path = labels_yaml_path
         self._det: YOLO | None = None
         self._cls: YOLO | None = None
         self._det_lock = Lock()
         self._cls_lock = Lock()
 
+    def _labels_path(self) -> Path:
+        raw = self._labels_yaml_path
+        if raw is not None and str(raw).strip():
+            return Path(str(raw).strip()).expanduser()
+        return Path(ba.CONSUMABLE_CLASSIFIER_LABELS_YAML_PATH).expanduser()
+
     def effective_candidate_consumables(self, requested: list[str]) -> list[str]:
         """请求体中的耗材子集；未提供（缺省或仅空白）时先用 ``consumable_classifier_labels.yaml`` 的 ``names``，无有效 YAML 则分类模型类名。"""
         out: list[str] = []
@@ -382,7 +388,7 @@ class ConsumableVisionAlgorithmService:
         if out:
             return out
 
-        yaml_path = Path(self._s.consumable_classifier_labels_yaml_path).expanduser()
+        yaml_path = self._labels_path()
         if yaml_path.is_file():
             ylist = list_sorted_class_names_from_yaml(yaml_path)
             if ylist:
@@ -404,7 +410,7 @@ class ConsumableVisionAlgorithmService:
         if not candidates_norm:
             return {}
 
-        yaml_path = Path(self._s.consumable_classifier_labels_yaml_path).expanduser()
+        yaml_path = self._labels_path()
         yaml_map: dict[str, str] = {}
         if yaml_path.is_file():
             try:
@@ -420,19 +426,14 @@ class ConsumableVisionAlgorithmService:
         return out
 
     def _det_weights(self) -> Path | None:
-        raw = (self._s.hand_detection_weights or "").strip()
+        raw = (ba.HAND_DETECTION_WEIGHTS or "").strip()
         if not raw:
             return None
         p = Path(raw).expanduser()
         return p if p.is_file() else None
 
     def _cls_weights(self) -> Path:
-        raw = (self._s.consumable_classifier_weights or "").strip()
-        if not raw:
-            raise ModelNotConfiguredError(
-                "未配置耗材分类权重。请设置 CONSUMABLE_CLASSIFIER_WEIGHTS。"
-            )
-        p = Path(raw).expanduser().resolve()
+        p = Path(ba.CONSUMABLE_CLASSIFIER_WEIGHTS).expanduser().resolve()
         if not p.is_file():
             raise ModelNotConfiguredError(f"耗材分类权重不存在: {p}")
         return p
@@ -468,7 +469,7 @@ class ConsumableVisionAlgorithmService:
         imgsz_det: int,
     ) -> np.ndarray | None:
         h, w = frame.shape[:2]
-        device = resolve_inference_device(self._s.hand_detection_device)
+        device = resolve_inference_device(ba.HAND_DETECTION_DEVICE)
         results = det_model.predict(
             frame,
             conf=det_conf,
@@ -499,28 +500,28 @@ class ConsumableVisionAlgorithmService:
             crop = self.hand_crop(
                 frame,
                 det_model,
-                det_conf=self._s.hand_detection_conf,
-                pad_ratio=self._s.hand_detection_pad_ratio,
-                min_crop_px=self._s.hand_detection_min_crop_px,
-                imgsz_det=self._s.hand_detection_imgsz,
+                det_conf=ba.HAND_DETECTION_CONF,
+                pad_ratio=ba.HAND_DETECTION_PAD_RATIO,
+                min_crop_px=ba.HAND_DETECTION_MIN_CROP_PX,
+                imgsz_det=ba.HAND_DETECTION_IMGSZ,
             )
             if crop is None:
                 return None
         else:
             crop = frame
 
-        device = resolve_inference_device(self._s.consumable_classifier_device)
+        device = resolve_inference_device(ba.CONSUMABLE_CLASSIFIER_DEVICE)
         try:
             r = cls_model.predict(
                 crop,
-                imgsz=self._s.consumable_classifier_imgsz,
+                imgsz=ba.CONSUMABLE_CLASSIFIER_IMGSZ,
                 device=device,
                 verbose=False,
             )
         except Exception as exc:
             raise PredictionError(f"耗材分类推理失败: {exc}") from exc
 
-        yp = Path(self._s.consumable_classifier_labels_yaml_path).expanduser()
+        yp = self._labels_path()
         if yp.is_file():
             st = yp.stat()
             index_to_label_id = _cached_index_to_label_id(
@@ -537,7 +538,7 @@ class ConsumableVisionAlgorithmService:
         )
         if snap is None:
             return None
-        if snap.t1_conf < self._s.consumable_min_cls_confidence:
+        if snap.t1_conf < ba.CONSUMABLE_MIN_CLS_CONFIDENCE:
             return None
         pname = snap.t1_name
         if not pname: