Add support for multimodal input

[Kludex]

The idea of this PR is to be able to send audio and images on the user prompt.

API

The proposed API involves changing agent.run():

import requests
from rich.pretty import pprint

from pydantic_ai import Agent
from pydantic_ai.messages import ImageUrl

image_url = 'https://goo.gle/instrument-img'

agent = Agent(model='google-gla:gemini-2.0-flash-exp')

output = agent.run_sync(["What's in the image?", ImageUrl(url=image_url)])
pprint(output)

You can also send multiple images from different sources:

from pathlib import Path

from rich.pretty import pprint

from pydantic_ai import Agent
from pydantic_ai.messages import BinaryContent, ImageUrl

image_url = 'https://goo.gle/instrument-img'
image_bytes = Path('docs/img/tree.png').read_bytes()

agent = Agent(model='google-gla:gemini-2.0-flash-exp')

output = agent.run_sync(
    [
        "What's in the image?",
        ImageUrl(url=image_url),
        BinaryContent(data=image_bytes, media_type='image/png'),
    ]
)
pprint(output)

The types added in this PR are:

• AudioUrl
• ImageUrl
• BinaryContent

Feedback welcome.

[github-actions]

Docs Preview

commit:	ef833f8
Preview URL:	https://0c6f3663-pydantic-ai-previews.pydantic.workers.dev

[Kludex]

Please follow #971

I couldn't fix the test suite, so I started from scratch, and I found out the issue, and I'm already adding tests over there.