CLoG: Benchmarking Continual Learning of Image Generation Models