Added support for google specific arguments for video analysis

[Sumered]

Added support for Google models specific arguments when processing video, those arguments are:

• media_resolution, which is model setting, settings this to LOW instead of default HIGH results in ~3x lower amount of input tokens consumed for video input.
• fps, which is video specific settings, set by setting vendor_metadata in FileUrl/BinaryContent, controls frame sampling. Default is 1.0 for Google models, setting this to lower value decrease amount of input video tokens, setting it to higher value increase analysis quality in highly dynamic videos.
• start_offset, which is video specific settings, set by setting vendor_metadata in FileUrl/BinaryContent, controls start offset of video. Useful for capping token consumption per video. According to docs it needs to contain s at the end, ex. 300s
• end_offset, which is video specific settings, set by setting vendor_metadata in FileUrl/BinaryContent, controls end offset of video. Useful for capping token consumption per video.

Official Google docs for those new arguments:
https://ai.google.dev/gemini-api/docs/video-understanding

[Sumered]

I'm still not exactly sure why are those tests failing, so I would be thankful if you could provide me with an explanation on what to do with those?

[DouweM]

Would this work?

Suggested change

	inline_data_dict['video_metadata'] = video_metadata # type: ignore
	inline_data_dict['video_metadata'] = item.vendor_metadata # type: ignore

[DouweM]

I'd prefer something like this:

Suggested change

	vendor_metadata: dict[str, Any] | None = None
	"""The vendor specific metadata for the file.
	Currently supports only those keys:
	fps: float,
	start_offset: str (ex. 1800s),
	end_offset: str (ex. 1800s)
	And works only for google models for video analysis.
	"""
	vendor_metadata: dict[str, Any] | None = None
	"""Vendor-specific metadata for the file.
	Supported by:
	- `GoogleModel`: `VideoUrl.vendor_metadata` is used as `video_metadata`: https://ai.google.dev/gemini-api/docs/video-understanding#customize-video-processing
	"""

[DouweM]

The test failure suggests that with the genai update, the previously recorded cassettes don't match anymore because post != POST... Can you try replacing method: POST with method: post in all the tests/models/cassettes/test_google/*.yaml files?