Terminating a running background job

When to use

• Cancelling a runaway build or long process spawned by shell::exec_bg.
• Cleaning up before re-issuing a corrected command.
• Reaping at the end of an orchestration before unregistering the worker.

Notes

• There is no signal field on the wire. Host kills go through tokio's Child::start_kill, a hard SIGKILL on Unix.
• A non-running job returns killed: false with reason: "not running". That is not an error.
• Sandbox-backed jobs cannot be hard-killed because sandbox::exec has no cancel hook. The record flips to killed and finished_at_ms is stamped immediately so shell::status and shell::list reflect cancellation, but the in-VM process keeps running until its timeout_ms expires. The response carries a reason explaining this.
• For real cancellation of a sandbox job, set a tight timeout_ms on the original exec_bg, or call sandbox::stop to tear down the VM.
• not_found (the trigger Err) means the job_id either never existed or aged out of retention.