Request to add SIBench evaluation code

[song2yu]

Added evaluation code for the SIBench paper: "How Far are VLMs from Visual Spatial Intelligence? A Benchmark-Driven Perspective https://arxiv.org/abs/2509.18905 ".

1. Includes inference_mixed.py to support mixed inference for both images and videos.

2. Includes SIBench.py for processing the SIBenchmark.

3. Introduced a new MixedOutput format.

4. Added post-processing support for the MixedOutput format in run.py.

[tonysy]

Please fix the lint issue

[song2yu]

Thanks for the reminder. I have already checked the code according to the development guide pre-commit run --all-files, and now shows no formatting errors.

[mzr1996]

Hello, I have checked the PR. Do we need to add an extra inference script? If it has video, I think directly using inference_video is enough.

[song2yu]

Thank you for your response. Since SIBench contains mixed data types—including single images, multiple images, and videos—a separate Inference_mixed.py is required to handle them respectively. Therefore, we recommend keeping it.