Improving Image Captioning Descriptiveness by Ranking and LLM-based Fusion