Proof Artifact Co-training for Theorem Proving with Language Models