Text this: Leveraging modality‐specific and shared features for RGB‐T salient object detection