tsdb: add an hourly rollup tier for the cluster table (tsdb_metrics_cluster_hour)

[renecannao]

PR #6034 replicates peers' raw TSDB samples into tsdb_metrics_cluster on the leader, with no rollup tier — v1 replicates raw only.

Measured cost (2026-08-13 lab run, 417-series real fixture): the cluster tier is ~104 B/row and a leader carries roughly 7.8 GB total (~2.3x a follower) at the 24h/14d profile, of which ~4.5 GB is cluster raw. Long-horizon node= queries are raw scans, and cluster raw retention must stay short (now 1 day) purely to bound size — which means the cluster view has no long-horizon data at all.

Design sketch (from the spec's follow-ups):

1. Leader downsamples its own tsdb_metrics_cluster into tsdb_metrics_cluster_hour keyed by (node, bucket, metric_name, labels), with its own retention variable. Self-sufficient, no protocol change.
2. Later: pull peers' tsdb_metrics_hour directly — the only way to obtain cluster history older than the backfill horizon at the moment a node first becomes leader.

Test bed exists: test/tsdb-lab/expand.py --nodes N produces a 14-day cluster table on demand.

Needs a design round (spec -> plan) before implementation.

[coderabbitai]

🔗 Related PRs

#6034 - Cluster leader election + leader-side TSDB stats aggregation (cluster as a single entity, phase 1) [open]
#6037 - Cluster stats aggregation: leader replicates peers' TSDB via pull+watermark [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!