Towards Omnimodal Expressions and Reasoning in Referring Audio-Visual Segmentation