Include time_out in the answer_generator run method

[rolandtannous]

Currently the OpenAiAnswerGenerator class doesn't accept timeout as a parameter, if someone is trying to use it together with a retriever in a pipeline.run() call.

The timeout parameter appears to be only malleable from a standalone generator.predict() call of the OpenAIAnswerGenerator. It would be great if we can pass the timeout parameter when initializing the OpenAIAnswerGenerator or from inside pipeline.run as a parameter.

Additional context
Appears related to this other feature request #3797

This came up after I started receiving timeout from the openAi server. It was working fine but then started timing out. It has been mentioned in this discord thread

[danielbichuetti]

As mentioned in Discord, I think that for timeout it doesn't make sense allowing to change it on each Pipeline run. It's not usual at least.

About the #3797, I can imagine some scenarios where it would be useful to set up max_tokens in different requests.

[bogdankostic]

In #3886 we added the environment variables HAYSTACK_REMOTE_API_BACKOFF_SEC and HAYSTACK_REMOTE_API_MAX_RETRIES which are currently only used in PromptNode. We should definitely use them also in the @retry_with_exponential_backoff decorator of the embed method of _OpenAIEmbeddingEncoder and _CohereEmbeddingEncoder and predict method of OpenAIAnswerGenerator.

Would someone of you be interested in creating a PR for this? :)

[danielbichuetti]

@bogdankostic I can proceed. Do you want the same pattern adopted in the PromptNode, right?! I'll commit next hours.