Rethinking Class-incremental Learning in the Era of Large Pre-trained Models via Test-Time Adaptation