Support dspy.Tool as input field type and dspy.ToolCall as output field type

[chenmoneygithub]

The main goal is making dspy.Tool a valid input type with special format so that users can easily use dspy.Tool without using dspy.ReAct. dspy.ToolCall is just a thing wrapper over tool name and tool args to simplify the output field definition involving tool calls.

Sample usage:

import dspy

dspy.configure(
    lm=dspy.LM("openai/gpt-4o-mini", cache=False),
    # lm=dspy.LM("anthropic/claude-3-5-sonnet-20240620", cache=False),
)


class MySignature(dspy.Signature):
    question: str = dspy.InputField(description="The question to answer")
    tools: list[dspy.Tool] = dspy.InputField(description="The tools to use")
    answer: str = dspy.OutputField(description="The answer to the question, if no more tool calls are needed")
    tool_call: dspy.ToolCall = dspy.OutputField(description="The tool call information, including name and arguments")


def get_weather(city: str) -> str:
    """Get the weather for a city"""
    return f"The weather in {city} is sunny"


tools = [dspy.Tool(get_weather)]
tools_dict = {t.name: t for t in tools}

predict = dspy.Predict(MySignature)

result = predict(question="What is the weather in Paris?", tools=tools)
if result.tool_call:
    print(f"Executing tool: {result.tool_call.name}")
    print(f"Tool result: {tools_dict[result.tool_call.name](**result.tool_call.args)}")

dspy.inspect_history()

With sample output (not including history):

Executing tool: get_weather
Tool result: The weather in Paris is sunny

[chenmoneygithub]

@okhat @TomeHirata I am going to add some adapter change to this PR, will ping you for review after that is done.