Reliability

• Restart a hung server: /healthz and a systemd watchdog
• Don't stay down after a background service fails
• Stop the biggest stream when the disk runs low
• Pause a remux's ffmpeg when it is far ahead of the player