PLEX: Perturbation-free Local Explanations for LLM-Based Text Classification