This PR adds a comprehensive model card for the dUltra model. It includes:
- Relevant metadata (`pipeline_tag` and `library_name`).
- A link to the paper: [dUltra: Ultra-Fast Diffusion Language Models via Reinforcement Learning](https://huggingface.co/papers/2512.21446).
- A link to the official GitHub repository.
- A brief summary of the model's architecture and performance.
- A sample usage code snippet extracted from the GitHub README.
- The BibTeX citation for the paper.
