You don't need to match the driver version between the host and the container. Actually, you shouldn't include any driver file inside the container.
All the user-level driver-files required for execution are mounted when the container is started using a volume. This way you can deploy the same container on any machine with NVIDIA drivers installed.
All the user-level driver-files required for execution are mounted when the container is started using a volume. This way you can deploy the same container on any machine with NVIDIA drivers installed.
We have more details on our wiki: https://github.com/NVIDIA/nvidia-docker/wiki/Internals
Concerning your last question: I don't have any information on this topic, but anyway it would not really impact nvidia-docker.