Prevent overlapping runs and schedule the nightly backup

locks.py plus the schedule, so it runs unattended without colliding with itself.

• A flock mutex stops two bur runs overlapping. A contended lock is a clean skip: log "run in progress" and exit 0, so it never leaks a false failure into the alert path. The lock is released by the kernel on a crash. It does not coordinate with deploy migrations (those run on app-container boot), which are left to pg_dump's MVCC-snapshot safety and per-database isolation. Don't claim flock guards migrations.
• One external cron does the scheduling: a systemd timer or cron entry running flock -n /var/lock/bur.lock bur backup --due. Discord reports each run. An optional bur install-timer can print or scaffold a unit, but it does not manage cron dynamically.

Acceptance

• a second concurrent run skips cleanly (exit 0) rather than alerting.
• the nightly --due run fires on schedule, and a failed run notifies via Discord.