Understanding why fixed memory limits hinder long-range comprehension in LLMs and how delta-mem techniques can enhance context retention and performance