Better audio representations are more brain-like: linking model-brain alignment with performance in downstream auditory tasks