CAMEL: Continuous Action Masking Enabled by Large Language Models for Reinforcement Learning