Exploring Lightweight Federated Learning for Distributed Load Forecasting