Upgrade to R 3.5.3

[RolandASc]

Dependency upgrade, i.e. specifically in the Dockerfile. At the same time, possibly check / review the handling of dependencies for guaranteed reproducibility and allow alignment of development versions of packages with what would be built / deployed through the Dockerfile.

(See in particular https://hub.docker.com/r/rocker/r-ver/dockerfile and the use of BUILD_DATE and MRAN.)

[riccardoporreca]

@RolandASc, given R 3.6.0 is out, shall we then directly upgrade to 3.5.3, which will be soon available with fixed MRAN build date?

[RolandASc]

Yes, makes sense, let's do that

[riccardoporreca]

Dockerfile considerations

Review and rethink the existing dependencies and their installation

• Required R package dependencies can be installed dynamically using remotes::install_local()
  • Still, there is the benefit in installing major, fixed depedencies like shiny and dplyr in a layer before the COPY to leverage caching and shared layers among different builds.
• Existing apt-get installations to be reviewed (not all might be needed). See below for some specific coments
• Do not rely on texlive-full, which is huge. Instead, use the portable, light-weight TinyTeX. This can be done by extracting the relevant parts from rocker/verse/Dockerfile. Note however that automatic cropping of R plots via pdfcrop is not part of the default TinyTeX installation (yihui/tinytex#118). Besides pdfcrop, we want to make sure all needed LaTeX pkgs (e.g. fancyhdr) are installed in the Dockerfile. Otherwise they will be installed on-the fly and the first user generating the report will have to wait for a few minutes!
• Do not install pandoc through apt-get, but get a more recent recommended version following new instructions for Linux. Version-specific .deb installers available for each release, it might be good to use the same version as in rocker/rstudio:3.5.3.