OLMoASR: A series of open speech recognition models (with open datasets)