feat: support Tool classes

[mplemay]

Currently the only way to define "tools" is via the following design patterns:

1. The @agent.tool decorator is great for simple agents
2. Wrapping an existing function in a Tool allows for re-usability, etc. tools

While these method are alright, they don't necessarily allow for a tool to have ClassVars or parameters. Given the shortcomings of the current approaches, I think it would be valuable to add a third approach which looks something like the following:

Note: this is a super simple example, but hopefully it illustrates the point.

from dataclasses import dataclass
from typing import ClassVar
from pydantic_ai import Agent

@dataclass
class MyTool:
    sum: ClassVar[int] = 0
    multiplier: int

    async def __call__(self, random_number: int) -> int:
        self.sum = sum + (random_number * self.multiplier)
        return self.sum

agent_1 = Agent(
    tools=[
        MyTool(multiplier=2)
    ],
    ...
)