S-Prompts Learning with Pre-trained Transformers: An Occam's Razor for Domain Incremental Learning