Advancing Multimodal Idiomaticity Representation