Update consumable pipeline, client API docs, and deployment config

- Refine effective candidate consumables and classifier labels
- Adjust vision algorithm, TSV logging, and video session wiring
- Refresh client surgery HTTP contract doc and staging/video docs
- Update settings, docker-compose prod, tests, and uv.lock

Made-with: Cursor

.env.example

@@ -31,9 +31,8 @@ CONSUMABLE_CLASSIFIER_TOPK=5
 # CONSUMABLE_MIN_CLS_CONFIDENCE=0.5
 # 时间窗（秒）：窗内多次推理取众数后再走自动记账 / 待确认。
 # CONSUMABLE_VISION_WINDOW_SEC=15
-# 可选：Excel「商品名称」「产品编码」表；空则物品 id 用名称。
-# 开始手术请求体 candidate_consumables 缺省或 [] 时，优先用本表全部商品名参与推理；未配置则用分类模型全部类名。
-# CONSUMABLE_CATALOG_XLSX_PATH=/path/to/视频中的商品信息表.xlsx
+# 业务物品 id 与开录时「空 candidate」的类名全表：app/resources/consumable_classifier_labels.yaml（names + label_id；同名多规格为 id1/id2/...）。未设或空则默认该包内文件。
+# CONSUMABLE_CLASSIFIER_LABELS_YAML_PATH=/app/app/resources/consumable_classifier_labels.yaml
 # HAND_DETECTION_WEIGHTS=/absolute/path/to/hand_detect.pt
 # HAND_DETECTION_IMGSZ=640
 # HAND_DETECTION_CONF=0.25