Federated Self-Supervised Speech Representation Learning