This PR enriches the dataset card for PAC Bench by adding:
- A link to the associated paper: [PAC Bench: Do Foundation Models Understand Prerequisites for Executing Manipulation Policies?](https://huggingface.co/papers/2506.23725)
- A link to the project page: https://pacbench.github.io/
- The `task_categories` (robotics) and relevant `tags` (`benchmark`, `physical-reasoning`, `vlm`, `robot-manipulation`) in the metadata for better discoverability on the Hugging Face Hub.
- A descriptive summary of the dataset based on the paper's abstract.
