Bug: finish_reason isn't used in batch request processors

[RyanMarten]

this should be added to GenericResponse
the finish reason should also be mapped to the standard set with the following

from litellm.litellm_core_utils.core_helpers import map_finish_reason
...
        # Get stop reason
        finish_reason = raw_response.get("stop_reason", "unknown")
        finish_reason = map_finish_reason(finish_reason)

        return GenericResponse(
            response_message=response_message,
            response_errors=response_errors,
            raw_response=raw_response,
            raw_request=None,
            generic_request=generic_request,
            created_at=batch.created_at,
            finished_at=batch.finished_at,
            token_usage=token_usage,
            response_cost=cost,
            finish_reason=finish_reason,
        )

[RyanMarten]

fixing this for anthropic in #536
but gemini and openai also need to be updated.

this is so requests can be retried on invalid finish reasons