Add support to provide images to prompts for GPT vision. Not messages/content is an array and individual messages has type key now