Video multimodal class alongside Image, Audio, PDF

[Eigenwise]

I maintain atomic-agents, which builds on instructor. We just got a downstream PR (Eigenwise/atomic-agents#274) adding video input support, and the contributor had to fall back to raw {"type": "video_url", "video_url": {"url": ...}} dicts, because instructor has no Video class to reach for. More providers accept OpenAI-style video_url content parts these days (MiniMax-M3 in that PR, Qwen-VL via DashScope, etc.), so I figure this will keep coming up.

Raw dicts do work, since convert_contents passes them through untouched. But they lose everything that makes Image/Audio/PDF pleasant: typing, the from_url/from_path helpers, and provider-specific conversion living in one place.

Concrete scope I have in mind:

• A Video class in instructor/processing/multimodal.py following the existing Image/Audio/PDF pattern
• from_url(), from_path(), from_base64() constructors with MIME validation for common video formats
• to_openai() emitting an OpenAI-compatible video_url content part; provider converters that don't support video raise a clear error
• Optional params like fps/detail where providers accept them

Prior art: #1850 asked for this for Gemini, and #1851 attempted it but was closed unmerged.

Happy to contribute the PR if this direction sounds right.