Previously binfmt-support was installed as a dependency of
qemu-user-static, but lately it became optional:

https://salsa.debian.org/qemu-team/qemu/-/commit/3f2a696

As a consequence it's not installed anymore in the container, instead
systemd is installed.

More or less a revert of 2fa6a33f

The warning is displayed when user run bash, apt or apt-get. That
should be enough to cover interactive use-case and CI use-case.

At the moment, this image is identical to kali-last-release. It's there
for compatibility. Soon we'll deprecate it.

At a first glance, it might seem that this commit is just renaming
'distro' to 'image'. But there's a bit more to it.

Up to now, we've been building *one container image per distro*. The
argument for each script was the distro, and then for each distro
corresponded an image.  The image name was equal to the distro name, the
only special case being the distro 'kali-last-snapshot', mapped to the
image 'kali'.

With this commit, we reverse this logic: we build different images, and
it just happens that to each image corresponds a distro. The argument
for each script is now the image, and the script deduces the distro from
the image name, if needed. Actually, only the script build-rootfs.sh
needs to know the distro, in order to debootstrap. All the other scripts
don't care, they just deal with images.

Why this change? First, it makes things simpler. We don't need to have
an IMAGE variable in the conf files anymore, since IMAGE is now part of
the conf filename.

This commit also prepares fo...

Introduced in 424bee1c, it turns out
that it's not really needed anyway, as far as I can tell.

In any case, the fix is now part of deboostrap 1.0.125 that was just
released in Debian unstable, cf:

https://salsa.debian.org/installer-team/debootstrap/-/commit/f49f48af

Not needed when running on GitLab CI, but definitely needed for local
runs. Otherwise we'll leave manifests around, and in the next run, new
images will be added alongside old ones in the existing manifests,
instead of creating new manifests.

Another approach could be to create the manifest with 'create' instead
of always running 'create --amend', in docker-push.sh. I *guess* it
would delete an existing manifest, if any? But that would be so awkward,
as it's not possible to create an empty manifest. So when we loop on
architectures, we need a special case for the first iter of the loop,
where we do 'manifest create', then in the next iterations we'll do
'manifest create --amend'... So awkward.

Local manifests are a bit of a mysterious thing, because there's simply
no way to list them via the cli, so it's very easy to forget that it
exists: https://stackoverflow.com/a/67406040/776208

It fails local build of extra images, as the images where not pushed, I
guess:

    #3 ERROR: docker.io/kalilinux/kali-rolling:2021.08.24-amd64: not found

Just trying to make clean and readable logs

... and fail early if ever binfmt doesn't work

Embarrassing to say, but this loop never worked

It's not set when running locally rather than in GitLab CI, obviously.

Due to previous commit

Workaround debootstrap's occasional failure, plus nitpicks

Closes #37

See merge request !34

Do the things in the same order for every stage

This was needed to install pixz, however we don't need pixz anymore in
this stage since 37d2f61a

Also drop explicit install of tar, as it's already installed in
docker:latest anyway.

The script build-rootfs is called many times to build various kali
distros, and various architectures for each of those distros.

It happens, from time to time, that the debootstrap call fails:

    Building rootfs kali-last-snapshot/arm64
    E: Unable to execute target architecture

This particular error message comes from the tool arch-test (called from
within debootstrap), when it tries and fails to execute a test binary in
a foreign architecture.

The reason for this failure is still mysterious. It's not easily
reproduced. And when it fails to build the kali-last-snapshot distro,
it means that it succeeded building kali-dev and kali-rolling before.
Nothing changed in the build environment in between.

So this commit simply wraps debootstrap in a retry loop, hoping that
it's enough. In case of failure, we dump some information about the
environment, in case it can help to understant what's wrong.

Fixes: #37

Refactor, cleanup, fix everything related to multi-arch

Closes #36

See merge request !33