feat(tools): parameter descriptions via Annotated and docstring Args

Closes #270

What

Two mechanisms for parameter descriptions on @ai.tool functions:

1. Annotated[str, Field(description=...)] now survives into the schema — the fix is get_type_hints(fn, include_extras=True); pydantic carries the metadata through create_model().model_json_schema() unchanged.
2. A Google-style Args: docstring section fills descriptions for parameters without a Field description. Field wins when both exist.

Why

Descriptions are how the model judges which tool to call and what each argument means. Both idiomatic typed-Python patterns were silently dropped, so every tool built this way shipped undocumented arguments.

Notes

• No new dependencies (the parser is ~40 lines in agent.py, Google-style only).
• Tools using neither mechanism get byte-identical schemas; there is a regression test pinning that.
• The tool description itself stays the full docstring — trimming the Args: block from it felt like a separate behavior decision, happy to follow up if wanted.
• Every provider protocol embeds spec.params verbatim (openai/protocol.py, anthropic/protocol.py, ai_gateway/protocol.py), so descriptions reach all models with no per-provider changes.

Validation

• uv run pytest: 768 passed (8 new tests covering annotated fields, nested Annotated[...] | None unions keeping constraints + description, multiline Args entries, Field-over-docstring precedence, unknown-name tolerance, and the byte-identical regression)
• uv run ruff format --check / ruff check: clean
• uv run mypy: 35 errors, identical to baseline on main (all pre-existing, none in touched code)
• uv run ty check: clean