Add response interceptor

[the-gigi]

The idea is similar to the request interceptor, which allows users to register a hook that can modify the request just before it is sent.

The motivation is a bug in the OpenAI compatibility of the Gemini API, which can be worked around by modifying the response coming back from Gemini.

See https://discuss.ai.google.dev/t/openai-compatibility-bug-in-response-object/58484

@sashirestela you mentioned in this simple-openai issue that I should open a cleverclient issue, but them said that it would be more difficult:
sashirestela/simple-openai#248 (comment)

So, instead of opening an issue I'm starting this discussion to understand the difficulty.

[the-gigi]

@sashirestela Looking at the code I see that there are many senders that extend HttpSender and oveeride the sendRequest() method. They all call at some point response.body(). For example:

public class HttpAsyncCustomSender extends HttpSender {

    @Override
    public Object sendRequest(HttpClient httpClient, HttpRequest httpRequest, ReturnType returnType) {

        var httpResponseFuture = httpClient.sendAsync(httpRequest, BodyHandlers.ofString());

        return httpResponseFuture.thenApply(response -> {

            throwExceptionIfErrorIsPresent(response, null);

            logger.debug("Response : {}", response.body());

            return JsonUtil.jsonToObject(response.body(), returnType.getBaseClass());
        });
    }

}

It seems that it's possible to add the response interceptor as another argument to sendRequest, or add a constructor and instead of directly calling response.body() the sender can call requestInterceptor(request.body()) if provided and it will return the modified response.

Is the difficulty come from the fact that all the sender classes need to be touched? or is there some other issue I'm missing. The change itself seems simple and similar across all the classes.

[sashirestela]

@the-gigi response.body() is of a different type in each scenario, so it is necessary to analyze how to handle it in them and, yes, it involves touching several classes, so more testing, more effort.

On the other hand, I don't see a real value in implementing it now or giving it a high priority, considering that I am thinking of making several changes to this library to make it more efficient and have a wider scope of application.

[sashirestela]

Done