A collection of sparse autoencoders pretrained to extract interpretable features from protein language models (PLMs)