Automated grading of Linux/bash examinations using large language models: a four-level cognitive taxonomy approach — ThinkLLM