rebuild all vmfbs from stock torq-compiler main ae286409 (all PP-OCR fixes merged); board-verified 10/10 at both det sizes

remove ppocr_det_640x384.vmfb (superseded by 640x352)

README: detector second size is 640x352

add ppocr_det_640x352.vmfb (replaces 640x384; board-verified 10/10, det 214 ms)

ppocr_det_640x384.vmfb: rebuild with the ConvTranspose interleave fix (torq-compiler-dev#2226); board-verified 10/10

Document the 640x384 detector size

Add 640x384 detector (16:9 bucket, 640x360 rounded up to multiple of 32)

Upload README.md with huggingface_hub

Upload assets/sample_ocr.jpg with huggingface_hub

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Drop the dense document sample; the menu card is the demo sample

Upload assets/sample_ocr.jpg with huggingface_hub

Upload README.md with huggingface_hub

Upload samples/sample.jpg with huggingface_hub

Add PP-OCRv6-tiny Torq NPU models (det 800x608 + rec width buckets), ONNX references and sample

initial commit