FAST: Federated Active Learning with Foundation Models for Communication-efficient Sampling and Training