Transform and to text parts for Google models that don't support tools

[sidg1215]

Fixes #3149.

[DouweM]

Because the arguments and return value can be very long, I want those on a separate line.

Also because there could be parallel tool calls with the same name, we should include the tool call ID.

I'd also want to format this in a more special way so the model understands this is not actually a user/assistant "message".

So I suggest doing something similar to how we send text files to OpenAI:

pydantic-ai/pydantic_ai_slim/pydantic_ai/models/openai.py

Lines 1006 to 1014 in c27e5e4

	def _inline_text_file_part(text: str, *, media_type: str, identifier: str) -> ChatCompletionContentPartTextParam:
	text = '\n'.join(
	[
	f'-----BEGIN FILE id="{identifier}" type="{media_type}"-----',
	text,
	f'-----END FILE id="{identifier}"-----',
	]
	)
	return ChatCompletionContentPartTextParam(text=text, type='text')

And we're doing that for Google as well: #3269.

[DouweM]

We should also test the mapping of the tool result

[sidg1215]

@DouweM Is there a way I can generate a casette file (I was initially attempting to right my own, but after thinking about it that feels...wrong)? I wanted to test for this behavior via Agent.run.

[DouweM]

@sidg1215 If you run the test with uv run pytest <path>::<test> --record-mide=recreate, vcrpy will automatically record them for you :)