Stateless tier: metrics-server, HPA, PDBs, Grafana dashboards

Part of #241 (phase 2). **Blocked by #244**, which brings the cluster
up.

## Summary

Establish the redundant stateless-workload tier: replicas ≥2 with
horizontal autoscaling, disruption budgets, resource requests, and
dashboards.

## Scope

- Install **metrics-server** (prerequisite for HPA and `kubectl top`).
- Require **resource requests** on stateless workloads so HPA and the
scheduler have signal.
- Configure **HPA** and **PDBs** for the stateless tier (replicas ≥2).
- Add **Grafana dashboards** for cluster + workload health, feeding the
existing Grafana Cloud stack (reuse the Fleet/Loki pipelines from
#150/#151 rather than re-inventing).

## Acceptance criteria

- [ ] metrics-server running; `kubectl top` works
- [ ] Stateless workloads carry resource requests
- [ ] HPA scales the stateless tier; PDBs protect it during
drains/upgrades
- [ ] Grafana dashboards for cluster/workload health live in the
existing stack
- [ ] New files carry SPDX headers; pre-commit passes