feat: support creating tools from functions with model parameters

[puzzle-solver]

Feature description

I would want to be able to use tools with defined by functions with parameters typed as pydantic models.

Motivation

Example use case: creating a cart with a list of items

class Cart(BaseModel):
    products: list[Product]
    
class Product(BaseModel):
    product_id: str
    quantity: int = Field(default=1, gt=0)

Then, I would want to define my tool like this:

def create_cart(cart: Cart) -> bool:
    """Creates a cart with items that a user can order"""
    if validate(cart):
        return "success"
    else:
        return "cart creation failed"

The agent definition would be just Agent(llm=llm, prompt=prompt, tools=[create_cart]).

---

Right now, specifying the tools this way does not capture the schema of a Cart and is not reflected in JSON schema that is created and send to the API.

The workaround for now is to use a custom tool:

def on_create_cart_tool_call(products: list):
    cart = Cart(products=products)
    return create_cart(cart)

Tool(
        name="create_cart",
        description="Creates a cart with items that a user can order",
        parameters=Cart.model_json_schema(),
        on_tool_call=on_create_cart_tool_call,
    )

However, this is much less clean.

Additional context

No response