Update test_detection.py to match refactored detection.py API

[Copilot]

detection.py was refactored to consolidate _convert_onnx_detections and _convert_yolo_detections into a single _convert_detections(predictions, batch, letterbox, model_type) function, and parse_detections now expects category_label in each detection dict. The tests were left referencing the old API.

Changes

• Imports: Replace removed _convert_onnx_detections / _convert_yolo_detections with _convert_detections
• TestConvertOnnxDetections / TestConvertYoloDetections: Update all call sites to use the unified signature:

  # Before (functions no longer exist)
  convert_yolo_detections(preds, tensors, paths, frames, sizes, letterbox=False, model_type='yolov5')
  
  # After
  _convert_detections(preds, [tensors, paths, frames, sizes], letterbox=False, model_type='yolov5')

• _make_detection_result + inline detection dicts: Add category_label key to detection entries to satisfy parse_detections' direct dict access
• test_filepath_preserved: Fix stale assertion ('my_image.jpg') that was testing with a different path than what was in the batch; now constructs a distinct-path batch to properly validate preservation