Uncertainty Principle for Communication Compression in Distributed and Federated Learning and the Search for an Optimal Compressor