Regex matching bug in code block detection

[mdlinville]

Describe the bug:

Triple-backtick code fences are being collapsed into a single line in the output, like:

Basic usage: ``` import weave```

Discovered this bug when trying to regenerate the Weave reference docs, which use Lazydocs for the Python SDK.

Expected behaviour:

Multi-line output, like:
Basic usage:
```
import weave
````

Hypothesis about root cause:

_RE_TYPED_ARGSTART = re.compile(r"^([\w\[\]_]{1,}?)[ ]*?\((.*?)\):[ ]+(.{2,})", re.IGNORECASE)
_RE_ARGSTART = re.compile(r"^(.+):[ ]+(.{2,})$", re.IGNORECASE)

The _RE_ARGSTART regex matches ANY line with a colon followed by 2+ characters. When it processes "Basic usage:" followed by doctest code in an Examples section, it incorrectly treats it as an argument and applies the substitution pattern r"\1: \2" (line 588), which tries to wrap the text after the colon in backticks.
The theory is that this regex is too greedy and gets applied in section_block contexts (like Examples) where it shouldn't.