Optimizing Audio Augmentations for Contrastive Learning of Health-Related Acoustic Signals