Add support for LXC-based workers

GPU pass-through does not yet work with all hosts, see autopkgtest #3, #4 and #5.

As a fallback, we should therefore support LXC-based workers.

Challenges:

• /dev/kfd and /dev/dri need to be accessible within the container
• Access to dmesg and amdgpu_firmware_info (#6)
• debci uses different queues for different backends, implying that
  • All hosts of a particular GPU architecture must use the same queue
  • All tests of a particular GPU architecture must be scheduled to that queue, so debci-scheduler needs to be made aware of that.