add stop sequence config for openAI chat config model

[sebastienblanc]

As specified here in a chat completion request we should be able to pass a stop parameter containing a string or list of strings.
It's implemented in upstream langchain4j , this PR just add it to the config and the recorder.

[geoand]

It would be great if we could have some kind of test for this

[sebastienblanc]

It would be great if we could have some kind of test for this

I agree but honestly how can we test this without hitting a real openAI endpoint ?

[geoand]

Can't we use Wiremock and configure it to respond with what OpenAI responds with and then just validate that we sent what we expected to send?

[sebastienblanc]

Can't we use Wiremock and configure it to respond with what OpenAI responds with and then just validate that we sent what we expected to send?

I have been looking a bit around but tbh I have not a clear idea on how to write this test. You say you just want to test the request is well formed ?

[geoand]

My idea is to validate that the request is what you expect it to be and have the response be whatever OpenAI gives (which you would hardcode by just copying the response from a real OpenAI request)