v0.2.20

Fixed

• A supervised CLI's setsid()-detached descendant — outside the round leader's process group — is now killed at the hard-kill fallthrough instead of being left running, orphaned, after the round is torn down.
• A configured docker run/podman run command (including sudo/env-wrapped and global-flag forms) escapes killpg-based termination and could previously outlive the round silently. It's now detected, agent-runner warns loudly and emits a round_container_orphan_risk event that the hard-wall can't guarantee stopping it, and best-effort-stops the container via an injected --cidfile for the plain (unwrapped, unflagged) form. Full container lifecycle management remains a later-release item.

Notes

• Not breaking: no agent-runner migrate step, no systemd unit change.
• peek --json gains the new round_container_orphan_risk event kind; existing kinds unchanged.