Fix stale backup metric reporting

[Jorres]

backups_failed_count / backups_succeeded_count are now recomputed in watcher
cycle from the last completed backup's status, instead of being set in memory on
backup-completion events. TBWR operations stop retrying after their schedule is deleted.

Behavioral changes

• Fixes: alert firing forever for a deleted schedule
• Metric series disappear within one watcher cycle after a schedule is deleted or deactivated
• Series now exist (value 0) for every active schedule, including ones with no completed
  backups yet (this is actually cool, we needed to wait for the first run to see zeros)
• A cancelled backup no longer sets failed=1; only a last completed backup in ERROR does
• A TBWR whose schedule was deleted finishes as CANCELLED between attempts; an
  already-running attempt is left to finish (not aborted in the middle), on success backup still available
• Metric updates lag up to one watcher cycle instead of being immediate on completion.
• on request (without_schedule) metrics are totally unchanged