E2E tests

[marcopiraccini]

Summary

• Add examples/storage-db/ — a multi-tenant Postgres KV proxy that demonstrates the coordinator pattern end-to-end (3 pods + 1 coordinator + Valkey + Postgres).
• Add test:e2e script and a CI e2e job that spawns coordinator + pod processes against GitHub Actions' Valkey + Postgres services.
• Rewrite the "Why Kubernetes" section of coordinator-pattern.md as "Platform requirements", with explicit subsections for Kubernetes, ECS / Fargate, and Nomad / VMs / Compose. The library has no K8s-specific code; the doc previously suggested otherwise.

What the e2e test covers

• All 3 pods register in Valkey and show up in GET /pods.
• POST /tenants/:id picks a pod via pickAndRegister and returns the chosen memberId.
• PUT/GET round-trip routes stickily to the owning pod.
• least-loaded strategy spreads new tenants across pods.
• Orphan reassignment: SIGKILL the owning pod, wait > TTL, the next request reassigns the tenant to a live pod and serves it.

Notes

• MEMBER_TTL env var added to the storage-db pod so the e2e can exercise orphan reassignment in ~4s instead of waiting on the 30s default.
• The pod's PUT/GET/DELETE handlers lazily ensure the per-tenant Postgres pool on first hit, so an orphan-reassigned pod can serve a tenant it never saw a POST /tenants/:id for. This matches the "cold local share opens on first hit" behavior described in coordinator-pattern.md.
• CI matrix is Node 22, 24, 26 for both the unit and e2e jobs.