The current X11 backend sends the image over the wire, which is slow and inefficient. We should probably use the MIT-SHM strategy for this crate if it's available.

(Also, we probably shouldn't create a new image every time we run the set_buffer command, but I think we can take care of that when we solve #37)