feat: Add middleware to tools

[ElectricCookie]

I wanted to be able to have a way to inject some logic into the tool calling at runtime, (my use case was confirming the use of certain tools).

This PR adds the ability to inject Middleware into the agent loop to control the usage of tools.

Middleware can intercept the tool call, pre and post process it.

The implementation allows for chaining multiple middlware(s? :D).

[csells]

That is super interesting! Can I ask for an example and a new docs page, too? Thank you for this!

[csells]

That is super interesting! Can I ask for an example and a new docs page, too? Thank you for this!

[ElectricCookie]

I'm going to play with this in my app for a bit and see if the API I defined works out when using it. The most simple middlware example:

 final agent = ai.Agent(
    "provider:model",

    tools: [...],
    middleware: [
      FunctionToolMiddleware((toolPart, tool, next) async {
        print('Logging Middleware - toolPart: ${toolPart.toJson()}');
        print('Logging Middleware - Tool: ${tool?.name}');
        final result = await next();
        print('Logging Middleware - Result: ${result}');
        return result;
      }),
    ],
  );

FunctionToolMiddleware is just a helper class implementation of the abstract middlware class, to inline a middlware in a functional style. This might not always be the cleanest approach in case you have some dependency injection or complicated logic :) therefore the pattern also allows extending the middleware class.