This PR adds the `pipeline_tag: any-to-any` to the model card's YAML metadata. Since S1-Omni processes and generates multiple scientific modalities (including text, SMILES, protein sequences, material CIFs, spectra, and scientific images), `any-to-any` is the most accurate classification for this unified multimodal reasoning model. This will help users discover the model when filtering by task types on the Hugging Face Hub.
