Cut the container's reachability, not its processes

Security suspects a credential leaked for the billing container. They need it unreachable in the next minute and undisturbed for the next hour.

PATCH /api/v1/containers/{id}/proxy/permissions/state with {"enable_proxy": false}. Every service URL on that container now answers 403: terminal, files, the app on 8080, all of it. No rule can re-open it while the switch is off. Inside, nothing changed. The database is still open, the worker is mid-job, the memory is intact, the logs are still being written. Nobody got in, and nobody has to race a shutdown to preserve state.

The investigation happens with the container running and the world locked out. When it's over, the same PATCH with true puts it back.

On a normal server the options were pull the cable or pull the plug, and each one destroys something.

The kill switch doesn't kill anything. It closes the one door.