Combining Federated and Active Learning for Communication-efficient Distributed Failure Prediction in Aeronautics