v1.0.0

 ## Added

    - **Prometheus and Grafana monitoring**, deployed with the stack from `docker-compose.metrics.yml`.
      - Prometheus is at `/prometheus` behind basic auth, Grafana at `/grafana`. Both log in as `hermes` with the password from `.internal.env`.
      - Nine dashboards: Overview, Rounds, Checks, Minion fleet, Minion capacity, API and web, Postgres, RabbitMQ and Swarm nodes.
      - The backend, minions and frontend export their own metrics (rounds and their phases, check results and durations, queue wait, deadline misses, minion slot use). Exporters cover Postgres, RabbitMQ, Traefik, dockerd, node_exporter and cAdvisor.
      - Metrics endpoints are never routed publicly by Traefik.
    - **Choose what Reset Competition clears.** Per enabled criterion, clear its earned data (Uptime rounds or inject submissions) and its score adjustments, and optionally every team's variable values. Only clearing Uptime rounds pauses the Uptime Engine.
    - **Scoreboard failure reasons.** Hovering a Down cell shows the check's output.

 ## Changed

    - Minions wait a random 0 to 100 ms before running each check.
    - The swarm deploy runs 16 minions by default and updates two at a time.

 ## Fixed

    - Rounds could stop after a redeploy because a dead backend session still held the scheduler lock. The backend now waits up to 8 s for the scheduler to release it on shutdown, and Postgres drops a vanished client's session within about 35 s.
    - WinRM checks leaked idle connections, SMB checks could hang past their deadline after logging in, and MySQL checks logged in twice.