What Matters to You? Towards Visual Representation Alignment for Robot Learning