Allow REPO_DIR to be a non-existing folder by creating it and providing the user with permissions

[bollwyvl]

PR summary by Erik 2022-10-31

• Non-existing REPO_DIR folder
  If users provide a REPO_DIR as a build arg to a folder that doesn't exist, the root user will create it but leave the non-root user without permissions to access it. This PR resolves that and adds a test to verify it by adding some mkdir and chown statements.
• Dockerfile refactoring
  It includes basic refactoring of the Dockerfile to reduce the layers by coalescing ENV statements and making use of the recommended syntax with explicit = in ENV statements.
• Refined test documentation
  It updates a comment to clarify how the test suite works.

---

Updated

This PR (originally attempted to try out some alternatives to #974 and (#975) now:

• changes the target dir TBD (might not, if [MRG] Ensure REPO_DIR owned by NB_USER #975 changes are sufficient: @manics and @tomyun offered some additional approaches to be considered/discussed)

• adds some additional read-/write-ability checks to the custom target dir test case
  • in the process of figuring out where to land it...
    • expands the docs in conftest.py for providing extra arguments via extra-args.yml
• partially to offset layer addition for path checking, where possible...
  • updates all uses of ENV to the recommended = syntax, replacing use of the "alternative" (space-based) syntax:

    The alternative syntax is supported for backward compatibility, but discouraged for the reasons outlined above, and may be removed in a future release.

  • combines multiple consecutive ENV calls to a single, multi-line directive

---

original

• fixes REPO_DIR permission denied #974 (attempted)
• alternative to [MRG] Ensure REPO_DIR owned by NB_USER #975

Some guidance on where to test this would be helpful... looking through the examples, I didn't see a way to add cli args, e.g. a repo2docker-cli-args file, perhaps that would be useful? this could be piggybacked off any of the existing ones, then...

I found ☝️ and added some docs in conftest.py. Since the conda example was already testing repo-dir, i just added the test in that place.

[tomyun]

Perhaps we can remove -R as it's an empty directory just created.

[bollwyvl]

yeah, it probably doesn't magically do umask or anything (though maybe we should think about that)...

[bollwyvl]

will wait to get any other feedback, it's a lot of coal to burn!

[manics]

FYI you can use something like /usr/bin/install -d <directory> -m <mode> -o <owner> -g <group> to do everything in one go if you want, doesn't really matter though.

[yuvipanda]

Yeah, can we use this instead? if in bash always makes me feel like I'm too dumb to make it work

[consideRatio]

This PR as a whole LGTM, but let's adjust to this review comment and resolve the merge conflicts and we are good to go.

[consideRatio]

Resolved it by applying my own suggestion, based on /usr/bin/install -d <directory> -m <mode> -o <owner> -g <group>

[consideRatio]

I applied a suggestion and rebased this PR on the main branch to get the test suite to run.

I took a backup of the state before rebase in my fork's branch.

[consideRatio]

Tests pass, I'll make another rebase to get a bit cleaner git history before merge.