feat: add Wan2.2 Animate 2 video models

[Minamiyama]

No description provided.

[gemini-code-assist]

Code Review

This pull request adds support for the Wan2.2-Animate-2-14B and Wan2.2-Animate-2-14B-Distilled character animation models (family WanAnimate2) by introducing a video parameter to the image_to_video endpoint across the RESTful API, async/sync clients, and diffusers model backend. Feedback highlights critical issues: both RESTful clients need to read the video file from disk when a local path string is provided rather than uploading the path string itself; in diffusers.py, fps must be popped from generate_kwargs to prevent a downstream TypeError crash, the temporary file path should be assigned before writing to ensure cleanup on failure, and local file paths should be validated for existence.

[gemini-code-assist]

There are three issues in this block:

1. fps is retrieved using get but not popped from generate_kwargs. Since fps is not a valid parameter for the underlying WanAnimate2Pipeline.__call__ method, leaving it in generate_kwargs will cause a TypeError crash.
2. If f.write(video) fails, temp_video_path remains None, leaving an orphaned temporary file on disk. Setting temp_video_path before writing ensures it is cleaned up in the finally block.
3. If video is a string path, we should check if the file exists to fail fast with a clear error.

            fps = generate_kwargs.pop("fps", 24)
            temp_video_path = None
            try:
                if isinstance(video, bytes):
                    with tempfile.NamedTemporaryFile(suffix=".mp4", delete=False) as f:
                        temp_video_path = f.name
                        f.write(video)
                    video = temp_video_path
                elif isinstance(video, str):
                    if not os.path.exists(video):
                        raise FileNotFoundError(f"Driving video path does not exist: {video}")
                else:
                    raise TypeError("`video` must be video bytes or a local path")

References

1. Avoid adding unused parameters to launch or configuration arguments if they are not directly consumed, as they can leak through **kwargs into downstream model construction. Instead, clean up stale or internal fields and overwrite only the required parameters.