CovRL-Fuzz: Fuzzing JavaScript Interpreters with Coverage-Guided Reinforcement Learning for LLM-Based Mutation