You are correct: there is currently no conversion to a depth-image. If the device is stationary, that representation could be quite useful. With a moving device, perhaps not, especially when turning.
To do that, one would need to write a package similar to velodyne_pointcloud, but using a different class to unpack the raw data. The logic for unpacking the data is in rawdata.cc. Code to create a polar, cylindrical projection would be a little simpler (less trigonometry). Making a package like that is feasible, but would require significant time and effort. The existing code would definitely provide a useful starting point.
I maintain the Velodyne ROS driver, but have not had much time to devote to it lately. I will gladly provide technical advice if you want to work on it. There are more technical details than would be worthwhile to discuss here, so please open an enhancement issue, if you wish to pursue the idea.