Update metadata: add task category and remove arxiv tag

This PR cleans up the YAML metadata:
- Removes `arxiv:2608.27345` from the tags, as the arXiv ID should only be referenced in the Markdown content.
- Adds `task_categories: image-to-video` to help users find the dataset under the correct task domain.

No content changes are needed because the existing README already contains the paper link, code link, project page, usage instructions, citation, and license information.

README.md

@@ -1,12 +1,13 @@
 ---
-pretty_name: PAWBench
 size_categories:
-  - n<1K
+- n<1K
+pretty_name: PAWBench
+task_categories:
+- image-to-video
 tags:
-  - benchmark
-  - video-generation
-  - physical-reasoning
-  - arxiv:2608.27345
+- benchmark
+- video-generation
+- physical-reasoning
 ---
 
 # PAWBench
@@ -148,4 +149,4 @@ the evaluator source code, not automatically to this dataset's images, prompt
 material, or scene metadata. No separate license has been selected for this
 dataset repository. Do not infer permission to redistribute or create
 derivatives of its assets without permission from the project owners. Source
-and package provenance is recorded in `scenes.jsonl` and `manifest.json`.
+and package provenance is recorded in `scenes.jsonl` and `manifest.json`.