Implement `finish_reason` in API response

Created by: frankxu2004

🚀 Feature Request

Implement finish_reason as in the OpenAI API specification. Currently it's default to "length".

It is useful for saving generation times and generate only until needed. It is especially useful for interactive prompt-based NLP tasks.

Note that the finish_reason field in the response should be under response["choices"][0]["finish_reason"] instead of response["choices"][0]["logprobs"]["finish_reason"] as implemented now in https://github.com/facebookresearch/metaseq/blob/51871bd73cd04c038f239ea2a26db1d7f6b37927/metaseq/service/responses.py#L33