- Bump to Ubuntu 20.04. This brings in lots of newer packages,
  but particularly Python 3.8 (https://phabricator.wikimedia.org/T265957).
  Many new packages are 3.7+ only.
- Use upstream R deb packages (https://cran.r-project.org/bin/linux/ubuntu/),
  than those from Ubuntu. This is maintained by the same folks who
  do R in debian. This brings us R 4.x (https://phabricator.wikimedia.org/T268923),
  but more importantly enables binary package installation.
- Use packagemanager.rstudio.com to install R packages. This is a
  huge benefit - it can install any version of packges, and it has
  binary package builds for most of them! These binary packages are
  built against focal + the upstream R debs. See announcement info
  here: https://blog.rstudio.com/2020/07/01/announcing-public-package-manager/.
  This reduces the user image build time a *lot*
- Put R packages in /srv/r, to match /srv/paws for our venv. Users
  can more easily install R packages temporarily now, and we don't
  mess with things under `/usr`. It also simplifies our Dockerfile,
  reducing number of times we have to switch to root.
- Redirect most stdout from apt-get to /dev/null. We still get stderr,
  so no real functionality lost.
- Remove some explicit nbextension enable / install commands. These are
  no longer necessary.
- Specify a loose pin for notebook & jupyterlab versions.

Uses github.com/jupyterhub/jupyter-rsession-proxy/ to provide
RStudio :)

You can test this with:

$ docker build -t paws images/singleuser
$ docker run -p 8888:8888 paws jupyter notebook --ip=0.0.0.0

This should start a notebook server, and print a URL with a token
you can use to connect to the docker container & test the image.
Since this uses port forwarding, it'll most likely only work if you
are running docker locally (and not via docker-machine)