This PR adds the missing `task_categories` field to the dataset card metadata, categorizing the dataset under `image-to-video` to improve discoverability on the Hugging Face Hub. The dataset is used for training world models that predict future video frames, which fits this category.
