This PR improves the dataset card for MMTR-Bench with the following changes:
- Updated the `task_categories` to `image-text-to-text` for better discoverability.
- Added links to the paper ([Can MLLMs "Read" What is Missing?](https://huggingface.co/papers/2604.21277)) and the official GitHub repository in the header and citation.
- Corrected the "How to Use" code snippet to match the actual dataset configuration (`MMTR.jsonl` and `test` split) and updated the printed fields based on the schema provided in the GitHub README (e.g., using `context_img_paths`).
