Improved Docker image build / sandbox setup process

[KyeRussell]

I've been playing with Packj in a way that has necessitated a lot of Docker rebuilds. I eventually opted to try to speed the build process up a bit. These are the associated changes, if you are interested in taking them on.

• Dockerfile changes:
  • Changed step ordering to decrease number of step rebuilds required when common things (e.g. the codebase) change.
  • Replaced usage of the apt command-line too with apt-get. apt has an unstable interface and its authors recommend against using it in an automated fashion.
  • Use Docker BuildKit caching for OS-level (apt), Python, and Ruby dependency installation steps.
  • Optimised Node.js installation process such that Node.js can be installed alongside the rest of the OS-level dependencies.
  • Only go through expensive / time-consuming sandbox setup process if the sandbox-related files have actually changed. If the codebase changes without the sandbox files changing, the previous sandbox setup persists.
• Sandbox setup (setup.sh) changes:
  • Added set -uo pipefail and IFS=$'\n\t' ('Unofficial bash strict mode') per bash practice for bash scripts. Note that set -e (usually part of strict mode) was not easily doable here (without changing how the script works). To minimise changes, I just excluded it, but it'd be nice to be able to use it at some point in the future.
  • Passing in the -v flag will print output that was previously just written to a log file. I added this to make the Docker image creation process more verbose, per best practice as you typically won't get access to log files written during the image build process.
  • Fixed typo ("Clonning" -> "Cloning").
  • Fixed some cases where the script would incorrectly return a zero exit status, which would incorrectly imply to Docker that the script had exited successfully, when it had in fact failed.

[ashishbijlani]

Look good.