Fix/yolo getting started docs

[LIKHITHA-NAGRAJ]

This PR improves the Getting Started Python operator dataflow example with better visualization, stability, and real-time performance observability.
Changes

• Headless CI fix: Set QT_QPA_PLATFORM=offscreen before cv2 import so the example runs in GitHub Actions without a display
• Dynamic reshape: Changed hardcoded (480, 640, 3) to (-1, CAMERA_WIDTH, 3) so it works on any camera resolution
  Empty bbox fix: Changed np.array([]) to np.zeros((0, 6)) to prevent crash when no objects are detected
• Clean metadata: Passing {} instead of dora_event["metadata"] fixes the Arrow type conversion error
• Performance overlay: Added real-time FPS and AI latency display
  Readable labels: Added filled background rectangle behind label text
• Optimized inference: Using imgsz=320 reduces latency significantly

BEFORE

AFTER

[LIKHITHA-NAGRAJ]

I've implemented the headless display fix and dynamic image reshaping across plot.py, object_detection.py, and webcam.py.
26/27 checks are passing. The remaining failure in CI / CLI Test (ubuntu-latest) is an internal timeout in the receive_data_with_sleep node caused by telemetry connection failures to 127.0.0.1:4317 — this is unrelated to the YOLO example changes in this PR.
Local testing on Windows shows AI latency dropping from ~863ms to ~53–150ms after switching to imgsz=320. Screenshots attached.
Could a maintainer please review when available? Happy to make any requested changes.