The previous design required two passes of the 10-minute sweep before a
maintenance request was created, so an outage took at least ~10 extra
minutes (up to 20) to be reported.
Split the monitoring into two crons with distinct responsibilities:
- the 10-minute sweep only discovers outages, flags the KO streak via
``http_first_ko_at`` and auto-resolves recovered services;
- a new 2-minute confirmation cron re-checks only the flagged services
and creates a request once the service has been continuously KO for
``HTTP_KO_CONFIRMATION_DELAY`` (lowered from 5 to 2 minutes).
This keeps the transient-outage filter while removing the extra sweep
latency. Services already under an open request are skipped by the
confirmation cron and remain handled by the sweep.
