feat(py): Add support for dynamically defined tools

[huangjeff5]

• Add a tool() constructor.
• Creates a per-call child registry to hold any inline tools that aren't stored in the Registry.

Sample usage:

import asyncio
from genkit import Genkit, tool
from genkit.plugins.google_genai import GoogleAI

DB = {
    'u-001': {'plan': 'Pro',  'balance': 1240,  'txns': ['2026-05-01 Invoice #88 -$500', '2026-04-28 Refund ORD-441 +$120']},
    'u-002': {'plan': 'Free', 'balance': 42.5,  'txns': ['2026-05-03 Top-up +$50']},
}

ai = Genkit(plugins=[GoogleAI()])

async def chat(user_id: str, msg: str) -> str:
    u = DB[user_id]
    async def account() -> str:
        """This user's account: plan and balance."""
        return f'plan={u["plan"]} balance=${u["balance"]}'
    async def transactions() -> list[str]:
        """Recent transactions for this user."""
        return u['txns']
    return (await ai.generate(
        model='googleai/gemini-2.0-flash', prompt=msg,
        tools=[tool(account, name='account'), tool(transactions, name='transactions')],
    )).text

async def main():
    for uid in ('u-001', 'u-002'):
        print(f'--- {uid} ---')
        print(await chat(uid, 'Summarize my account and transactions.'))

asyncio.run(main())

[gemini-code-assist]

Code Review

This pull request introduces support for inline tools by scoping unregistered tool instances within temporary child registries during generation calls. It refactors the prompt execution pipeline to separate template rendering from generation options and adds a "resume" field to PromptConfig for enhanced tool response handling. Feedback was provided regarding a PEP 8 line length violation in _generate.py, suggesting a refactor for better readability.