Plex: Perturbation-Free Local Explanations for LLM-Based Text Classification