The Effect of Batch Size on Contrastive Self-Supervised Speech Representation Learning