add monitoring to dealbot ops dashboard for cleanup job

## Description

The piece_cleanup job (added in #346) deletes the oldest pieces from each SP once that SP exceeds MAX_DATASET_STORAGE_SIZE_BYTES. It already emits the signals we need (runs, deletions, runtime, no_progress events), but the ops dashboard does not surface them.

This issue extends the dashboard pattern from #89 (now closed) with cleanup-specific charts and alerts on BetterStack, so on-call can see a stuck or runaway cleanup loop without grepping logs.

## Impact

Today, a problem with cleanup only surfaces through downstream symptoms: storage growth, deal failures, or rising over-quota counts. Root cause is several layers removed by then. Direct visibility shrinks the gap from "alert fired" to "we know why".

## Proposed Approach

Add to the existing BetterStack dealbot ops dashboard.

Charts:
1. Pieces deleted per SP per hour
2. piece_cleanup_no_progress events per SP per hour
3. Cleanup job runtime distribution, with a marker at MAX_PIECE_CLEANUP_RUNTIME_SECONDS
4. Active over-quota SPs (gauge of SPs currently above MAX_DATASET_STORAGE_SIZE_BYTES)

Alerts (route to the same Slack channel as the existing #89 alerts):
- 3 consecutive piece_cleanup_no_progress events for the same SP (cleanup runs at 1/hour per SP by default, so this fires after ~3 hours stuck; per #346)
- Job runtime equals MAX_PIECE_CLEANUP_RUNTIME_SECONDS three times in any 6-hour window
- An SP stays over MAX_DATASET_STORAGE_SIZE_BYTES for more than 24 hours

## Acceptance

- All four charts live on the ops dashboard
- All three alerts wired and verified with a synthetic event
- Dashboard config committed or backed up per #89 conventions
- Runbook (#508) links to the dashboard for the relevant alert IDs