Adds roles/monitoring with monitor_prod.sh — silent cron watchdog that
checks hermes_infra services, restart flapping, failed tasks, brencher
liveness, disk and memory. Stdout empty when healthy; non-empty when any
check fails (delivered as a single Telegram alert by cron no_agent job).
Script is already running as cron job 'prod-monitor' (id 4ebfb853ea68)
every 5m against docker_socket_proxy:2375. Role currently only ships
the file; cron wiring on remote is the caller's responsibility until a
follow-up adds a systemd/cron task.