DRL-TinyEdge: Energy- and Latency-Aware Deep Reinforcement Learning for Adaptive TinyML at the 6G Edge | Synapse