Text this: Detecting Malicious Clients Based on Autoencoders in Federated Learning for Indoor Localization