fix(aiot): 降低YOLO置信度阈值(0.5→0.4)减少离岗误报

人员在岗但检测置信度低于0.5被过滤，导致算法误判"无人"触发离岗告警。
将默认阈值降至0.4，同步更新环境变量默认值。

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

config/settings.py

@@ -117,7 +117,7 @@ class InferenceConfig:
     input_width: int = 480
     input_height: int = 480
     batch_size: int = 1
-    conf_threshold: float = 0.5
+    conf_threshold: float = 0.4
     nms_threshold: float = 0.45
     device_id: int = 0
     fp16_mode: bool = True
@@ -273,7 +273,7 @@ class Settings:
             input_width=int(os.getenv("INPUT_WIDTH", "480")),
             input_height=int(os.getenv("INPUT_HEIGHT", "480")),
             batch_size=int(os.getenv("BATCH_SIZE", "4")),
-            conf_threshold=float(os.getenv("CONF_THRESHOLD", "0.5")),
+            conf_threshold=float(os.getenv("CONF_THRESHOLD", "0.4")),
             nms_threshold=float(os.getenv("NMS_THRESHOLD", "0.45")),
         )