We can report group termination status more accurately if we know that
the termination was requested by the engine. ExecContext() uses
SIGKILL to stop processes, which tends to result in a "killed" error
depending on who wins the errgroup race: this is expected, so it would
be best to report it as such.

Conversely, a successful exit status is unexpected if the engine did
not explicitly request termination.

The new Engine can allocate subnets and IPs because it introduces an
explicit abstraction for shared (allocatable) resources, and plugs it
into the CreateGroup API.

VMProvider and NetworkProvider have been abstracted behind interfaces
to allow for testing of the Engine itself.

Update module golang.org/x/sync to v0.11.0

See merge request !19

The request payload needs an update with Gitlab CI specific info.

Necessary for trixie and the apt_repository Ansible module.

Update module golang.org/x/sync to v0.10.0

See merge request !18

Also: add libpam-systemd (logind/dbus).

This error was preventing the root filesystem from being remounted
read-write.

There's no longer a need to update the script whenever the minor
Debian version changes, as it will automatically find the most recent
one from the Gitlab API.

Simplify the CI script by breaking out its components

See merge request !17

Autopkgtest is just a wrapper, it's simpler to use vmdb2 directly to
build the image as we want it.

The nbd module is not available in the gitlab runners docker env.