This PR improves the model card metadata and content by:
1. Adding the `library_name: transformers` metadata since this model natively supports loading via the `transformers` library using `trust_remote_code=True`.
2. Updating the `pipeline_tag` from `image-to-image` to `any-to-any` to represent its native unified multimodal understanding and generation capabilities (visual reasoning, text-to-image, image-to-image, etc.).
3. Integrating links to the Hugging Face Paper Page ([HunyuanImage 3.0 Technical Report](https://huggingface.co/papers/2509.23951)) in the badges and the introduction text.
Fix: Implement deepseek moe for moe_impl == "eager" to solve oom (#5)
- Update README huggingface and official links (44bd9b987cc6d51a225966792249c2ba825e7ef0)
- Update README_zh_CN huggingface and official links (183be7a35e01e85b97a2367f0e9b7bc664964710)
Co-authored-by: lijunzhe <tulvgengenr@users.noreply.huggingface.co>