fix: optimize diagnosis output format and fix inconsistency between fault_status and fault_type

1. 输出格式优化:
   - faults中移除冗余的bbox和confidence字段
   - summary中fault_count改名为detection_count

2. fault_status与fault_type逻辑统一:
   - normal状态时fault_type为'none'，不再使用'suspect'
   - abnormal状态时才进行故障类型分类(hotspot/local_overheating/overheat)

3. 修复usage_example.py图像通道问题:
   - YOLO检测使用彩色图，温度提取时再转灰度图

scripts/usage_example.py

@@ -52,10 +52,13 @@ def example3_temperature_extraction():
     detector = InfraredDetector("weights/best.pt")
 
     # 读取图像
-    gray_img = cv2.imread(SAMPLE_IMAGE, cv2.IMREAD_GRAYSCALE)
+    color_img = cv2.imread(SAMPLE_IMAGE, cv2.IMREAD_COLOR)
 
-    # 检测目标（复用已加载的图像）
-    detections = detector.detect(gray_img)
+    # 检测目标（彩色图）
+    detections = detector.detect(color_img)
+
+    # 转为灰度图用于温度提取
+    gray_img = cv2.cvtColor(color_img, cv2.COLOR_BGR2GRAY)
 
     # 提取第一个检测框的温度特征
     if detections: