Re-factor the run method of JobsRunnerAsync so prompt logic is pushed down to Interview level

[johnjosephhorton]

Re-factor this hacky bs:

        # TODO: move this down into Interview
        question_name_to_prompts = dict({})
        for result in valid_results:
            question_name = result["question_name"]
            question_name_to_prompts[question_name] = {
                "user_prompt": result["prompts"]["user_prompt"],
                "system_prompt": result["prompts"]["system_prompt"],
            }

        prompt_dictionary = {}
        for answer_key_name in answer_key_names:
            prompt_dictionary[
                answer_key_name + "_user_prompt"
            ] = question_name_to_prompts[answer_key_name]["user_prompt"]
            prompt_dictionary[
                answer_key_name + "_system_prompt"
            ] = question_name_to_prompts[answer_key_name]["system_prompt"]

        raw_model_results_dictionary = {}
        for result in valid_results:
            question_name = result["question_name"]
            raw_model_results_dictionary[
                question_name + "_raw_model_response"
            ] = result["raw_model_response"]