Text this: A Flexible Model Compression and Resource Allocation Scheme for Federated Learning