bug(sandbox): openshell-sandbox keeps running after SIGTERM, so every stop waits for the runtime's kill timeout #4635

Problem

When the container runtime sends SIGTERM to openshell-sandbox (PID 1 of the sandbox), it logs "received termination signal", then "workload frozen pending authenticated recovery", and keeps running. The runtime waits out its whole stop timeout and then SIGKILLs it.

Effect

Every openshell sandbox stop takes the full stop timeout (10 s here), and the workload never gets a clean shutdown.

Reproduce

v0.1.4, external driver for Apple container (runtime stop timeout 10 s):

openshell sandbox create --name rc --from alpine:3.22 --detach -- sleep 3600
time openshell sandbox stop rc   # ~10 s

Proposal

After a termination signal, terminate the workload and exit within a bounded grace period (or document which signal or RPC a driver should use for a prompt, clean stop).