eNODE: Energy-Efficient and Low-Latency Edge Inference and Training of Neural ODEs