Clean slate: 1h block sizes, remove backfill artifacts

- Changed all namespace block sizes to 1h (was 2h/12h/24h in manifests,
  30d+ in the live cluster due to backfill-era bufferPast hacks)
- Deleted entire backfill/ directory (scripts, pods, runbooks)
- Removed stale 05-m3coordinator.yaml (had backfill namespaces)
- Added 05-m3coordinator-deployment.yaml to kustomization
- Fixed init job health check (/health instead of /api/v1/services/m3db/health)
- Updated .env.example (removed Mimir credentials)
- Added 'Why Backfill Doesn't Work' section to README

.env.example

@@ -6,10 +6,6 @@
 M3DB_USERNAME=<your-m3db-username>
 M3DB_PASSWORD=<your-m3db-password>
 
-# Mimir (source for backfill)
-MIMIR_USERNAME=<your-mimir-username>
-MIMIR_PASSWORD=<your-mimir-password>
-
 # Grafana Admin
 GRAFANA_ADMIN_PASSWORD=<your-grafana-admin-password>