Ikiwiki won't render strings, that are marked as fuzzy. So we unfuzz the
strings, so the sugestions get rendered in the staging website.

Still we don't know how celery works exactly and it seems, like
commit_pending is not executed at all. Let's do it old style and use
cron for it. At least it should not break anything, as Weblate is using
lock files to control write access to the repository and Weblate is
happy, if nothing is up to commit.

On top of DRY'ing our code, my main goal is to repair a use case
that 7591f7ae broke: a Jenkins
slave (sic) that can both build & test Tails images.

I've added this packages 6.5 years ago via
a5b072bbd10cd83d6cd194bf1eba1014393b9333 in puppet-lizard-manifests.git, because
"It's needed to run the test suite of our Perl projects".

But:

 - I've never set up Jenkins jobs that run the test suite of our Perl projects.
 - This package has been orphaned (https://bugs.debian.org/946410).
 - Jenkins now has a TAP plugin (https://plugins.jenkins.io/tap),
   so if we ever set up Jenkins jobs for test suites that natively expose
   their output in the TAP format, we can use that plugin instead
   of converting TAP to JUnit XMP.

The internal update jobs of Weblate are handled by celery and not
anymore by cronjob. We can remove any usage of internal cronjobs.

The only needed change is that models.SubProject has renamed to
models.Component.

The archive will go under that directory.

We want to first give the Tails community a chance to validate that it's safe to
make this archive public.

We disable apache and cron jobs in order to upgrade database smoothly,
as per upstream's documentation.

  - The settings.py template has been upated to match upstream.

  - We have to revert commenting out of AUTH_USER_MODEL after database
    has been upgraded.

Upgrading to the Buster version did not fix the problem that I'm investigating.

I'm seeing a 1 byte difference in the SquashFS generated on my sid system and
the one generated on our Stretch isobuilders. This difference seems to be
related to /bin/ping's xattrs, which indeed the IUK generation process
messes up with. So let's try if having the same version of the attr
package on both sides helps.

We need to standardize on one single version of squashfs-tools to get
reproducible IUKs.

We need to standardize on one single version of squashfs-tools
(ideally 4.4 which builds smaller IUKs), that can run on Stretch,
Buster, and sid.

"dzil authordebs" does not install runtime dependencies.