Decentralised Learning with Random Features and Distributed Gradient Descent