Active Learning for Abstractive Text Summarization via LLM-Determined Curriculum and Certainty Gain Maximization F i n d i n g s - E M N L P