OCR images in the current note using on-device PaddleOCR via ONNX Runtime — no server, no API key. Recognized text is inserted inline below each image.