[tf-notebook] Use nvidia/cuda runtime image instead of nvidia/cuda devel image

[r2d4]

For GPU support in our tensorflow notebook image, we currently use the development base image for cuda.

Switching to the runtime image should give us much smaller image sizes with the same functionality.

From the dockerhub page:

runtime: extends the base image by adding all the shared libraries from the CUDA toolkit.
Use this image if you have a pre-built application using multiple CUDA libraries.
devel: extends the runtime image by adding the compiler toolchain, the debugging tools, the headers and the static libraries.

https://hub.docker.com/r/nvidia/cuda/

Sizes of base iamge
9.0-cudnn7-devel-ubuntu16.04: 2.584 GB
9.0-cudnn7-runtime-ubuntu16.04: 1.148 GB

Savings: 1.436 GB