Load a tiny video to make CI faster

[zucchini-nlp]

What does this PR do?

As discussed internally, let's load a smaller video here because the processor doesn't do video sampling during the test. The new tiny video consists of only 10 frames

Some values were updated in tests because the sampling logic depends on loaded video's length and FPS

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ydshieh]

OK, we will need to add this new url to the script. I can do that in my PR.

Otherwise just make sure the new values works for the new video that would be good

[zucchini-nlp]

Oke, will add url_to_local and merge. The new values are working for all models in this test case

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: glm4v, qwen2_5_omni, qwen2_5_vl, qwen2_vl, smolvlm