feat: react to guest shutdown events (graceful shutdown, guest side)

[Anamika1608]

What this does

• Adds the guest side of graceful shutdown. urunit now reacts when the runtime asks the guest to stop, instead of ignoring the request.
• Pairs with the urunc host side in feat(monitors): graceful shutdown over the control socket urunc-dev/urunc#869, which injects the shutdown event. This PR makes the guest act on it.
• Part of Explore options for graceful shutdown of Linux-based VMs urunc-dev/urunc#336.

How it works

• Opens the power-button input device (/dev/input/event* that reports KEY_POWER). QEMU and Cloud Hypervisor deliver a power-button press this way.
• Sets up a signalfd for SIGCHLD, SIGINT and SIGTERM, and calls reboot(CAD_OFF) so a guest Ctrl+Alt+Del (Firecracker on x86) arrives as SIGINT to PID 1.
• Runs a poll loop that reaps children and watches for a shutdown event.
• On a shutdown event, sends SIGTERM to the app's process group once, then keeps waiting for the app to exit.
• After the app exits, runs the existing sync, unmount and reboot steps.
• Creates /dev and mounts devtmpfs at start, because the guest initrd has none.

Design choices

• No internal timeout. The app gets SIGTERM and time to exit. The container manager's SIGKILL is the outer bound. This matches the host side.
• The exit path stays reboot(RESTART). On QEMU (-no-reboot) this is a clean stop.
• If no power-button device exists, urunit still runs; the SIGINT path keeps working.

Status

• Not yet validated end to end. The guest kernel in the current test images does not build the input and button drivers (CONFIG_INPUT_EVDEV, CONFIG_GPIO_PL061, CONFIG_KEYBOARD_GPIO), so the power button never reaches userspace. A guest kernel with these options is needed to test the full path.
• Firecracker on x86 (Ctrl+Alt+Del) is code only, since there is no x86 host here.

Testing

• Builds clean as static and static_debug under the strict flags (-Werror -Wextra -Wall -pedantic-errors).
• Reviewed for signal, fd and reaping correctness.
• e2e testing for the qemu vmm