This PR improves the model card by:
* Adding `pipeline_tag: image-text-to-text` to the metadata, which helps with discoverability for models that evaluate text-to-image generation.
* Adding `library_name: hpsv3` to correctly reflect the primary library used by the model for inference and training.
* Fixing the broken paper link in the "Introduction" section to point to the Hugging Face Papers page (`https://huggingface.co/papers/2508.03789`).
* Correcting the arXiv badge link to `https://arxiv.org/abs/2508.03789` and updating its text to include the arXiv ID for better clarity.
* Updating the "How COHP Works" section to reflect 4 rounds of iterative refinement and the corresponding adaptive strength.
* Correcting one of the email addresses in the "Support" section.
