utils: keep docstring continuation lines that contain a colon

[breken-ai]

When a function is passed as a tool, a wrapped argument description is cut short if a continuation line contains a colon or a parenthesis:

def get_page(url: str, date: str) -> str:
    """Fetch a page.

    Args:
        url: Page to fetch, for example
            https://example.com/docs
        date: Day to fetch, in the format
            YYYY-MM-DD (for example: 2024-01-31)
    """

The schema that chat(tools=[get_page]) sends today:

{"url":  {"type": "string", "description": "Page to fetch, for example"},
 "date": {"type": "string", "description": "Day to fetch, in the format"}}

_parse_docstring strips every line in the Args: section and then treats any line with a : (or a (...)) as a new argument. The continuation lines https://example.com/docs and YYYY-MM-DD (for example: 2024-01-31) become phantom arguments named https and YYYY-MM-DD. The real descriptions stop at the line before, so the model never sees the URL example or the date format. Continuation lines without a colon already work, and test_function_docstring_parsing covers them.

This change keeps each line's indentation for the Args: section. A line indented deeper than the current argument's line is treated as a continuation, even if it contains a colon, which is how Google-style docstrings wrap descriptions. Lines at the argument's own indentation are parsed as before, so the existing parenthesis and skewed-docstring cases don't change.

Verification

• New test_docstring_continuation_lines_with_colons fails on main ('Page to fetch, for example') and passes with this change.
• End to end through Client.chat(tools=[get_page]) with an httpx.MockTransport: the request now carries "Page to fetch, for example https://example.com/docs" and "Day to fetch, in the format YYYY-MM-DD (for example: 2024-01-31)".
• pytest ollama tests passes (103 tests), and ruff format --check and ruff check are clean.

Built by breken, your AI support engineer - breken.ai - this one's on us.