Update README.md

Update README.md

Update README: repo is now self-contained with images, simplify Kaggle guide accordingly

Add the 98 real images the benchmark was computed from, so it can be run/demoed directly from this repo

Add real 3-engine (Tesseract/EasyOCR/PaddleOCR) benchmark results, ground-truth manifest, and Kaggle reproduction guide

initial commit