Build a JSON schema from a function's signature

[rlouf]

A common use case for JSON-guided generation is to extract values that are then passed to a function (see autonomous agents applications. This PR adds the possibility to extract the JSON specification from the signature of a function. Under the hood we create a Pydantic model from the extract argument names and type annotation.

The use interface is the following:

from outlines import models
from outlines.text import generate

def do_something_with_result(foo: str, bar: List[int]):
    pass

model = models.transformers("mistalai/Mistral-7b")

generator = generate.json(model, do_something_with_result)
args = generator("Generate arguments for the function")
do_something_with_result(**args)

Closes #350. Closes #351

TODO

• Write documentation (waiting for Use MkDocs for the documentation #331 to be merged)
• Return a BaseModel instance when input schema object is a Pydantic model