Bare-metal workers via PXE network boot

It would be nice to be able to run tests on bare metal. We could take an approach quite similar to the qemu backend, where the host creates and image and starts a virtual machine with the image to execute the tests. With PXE network boot, the conceptual difference would merely be that both the host and the guest would be two real machines communicating over the network rather than one real and one virtual machine.

Using bare metal sidesteps problems with qemu, where pcie passthrough might not be possible for certain hardware (APUs, SoCs, and single-GPU systems that don't have a working pcie reset). It also sidesteps problems with podman (inability to test driver updates, or kernel and GPU state from earlier tests affecting later tests). Support bare metal may also have value when working with complex, multi-GPU systems. We may be able to make a more convincing case to have whole servers donated to the Debian CI if we can promise that the test environment exactly matches the real-world usage environment.

In any case, this is clearly a lot of work and it's not a high-priority thing, but I wanted to ensure there's some visibility for it.