`sensor_msgs_py.point_cloud2` functions treat numpy arrays as having shape (width, height)

I'm not sure if this is really a problem or just a different convention but it was something that surprised me that I had to work around. I see that for organized clouds both the read_points and create_cloud functions have logic that treat the input/output ndarray of having a shape where the cloud width is the first index (rows), and the cloud height is the second (columns). I suppose it works if you always follow that convention but I think most other code will treat it with the opposite convention and some axes swapping is necessary.

I think pretty much only the bits I linked would have to be changed to make the switch. @Flova what are your thoughts as the original PR author?