Verifier-Based Reinforcement Fine-Tuning of Reasoning Models for Thermal Energy Storage Control