Training Vision Transformers in Federated Learning with Limited Edge-Device Resources