QFA2SR: Query-Free Adversarial Transfer Attacks to Speaker Recognition Systems