This fixes a regression introduced by
commit:787eb59e, which renamed the artifacts but
did not adjust the code that retrieves them from the build VM.

The basebox currently used by most our branches is now 4 months old,
so every build on Jenkins starts by deleting it… and then recreates
it from scratch. Six months is more in line with the delay between
our major releases these days.

This reverts commit fc1bba10.

Rake setup_environment: display a more helpful error message when the 'origin' remote does not point to the official Tails Git repository

This task calls git_base_branch_head() which relies on the fact 'origin'
points to our official repo.

The output of `free' has changed in Stretch.

The output of `free' has changed in Stretch, and we now require Stretch or newer
as the host system. It seems that we didn't notice this problem earlier as this
check is performed only when the build VM is not running yet, which rarely
happens in practice at this point.

Refs: #12599

Refs: #12575, #12531, #12576

Array#sort! wants a <=>-comparison (i.e. -1, 0, 1), not a boolean one.

Will-fix: #12575

There's some overhead with each ssh call that we might as well avoid,
and each scp invokation prints a warning ("Warning: Permanently added
[...] to the list of known hosts") that is nicer to just see once.

I added the arguments to these in anticipation of them being useful
for other boxes. That didn't turn out to be the case, so let's remove
them.

* Committing anything to vagrant/definitions/tails-builder means there
  is a new version of the builder VM (and it will automatically be
  generated next Tails build)
* Separately track the time-based snapshot serials used to generate
  the builder

These changes makes it extremely convenient to release new boxes (just
commit) and avoids issues where commits made to
vagrant/definitions/tails-builder may affect *some* users base box
generation, but not others, depending on when they Git fetch (see
ticket #12409, comment 32).

Will-fix: #12409

I.e. the one we have SSH access to through Vagrant and hence what
run_vagrant_ssh() targets -- the moved code would fail for other
domains!

Otherwise we might experience corruption.

... instead of downloading it. While there are many nice
advantages (and a few disadvantages) the main reason for this is to
have one less binary blob as input in the build process.

Will-fix: #12409

It returns the empty list, not nil, when there's no match.

Refs: #12409

I apparently messed up a lot (typos, bad decisions) in the previous
four commits.

I.e. we do not require building from detached head any more.

This way changing e.g. args to `run_test_suite` or cucumber (like
`--tag`) can be controlled from the Rakefile, which gives us a lot
more flexibility.

Apparently Jenkins sets it like so, so when we were checking out that
ref, we end up in detached head.

Apparently that automatically happens when sypplying a `desc`.

With only suppress the output of ssh itself, like "Connection to
1.2.3.4 closed" -- the output of commands are unaffected.

We want this on Jenkins.

It was a misunderstanding to remove it.

Refs: #11972