chore: Prometheus /metrics endpoint + Grafana dashboard

[qvkare]

Problem

No runtime metrics: we can’t observe latency spikes or RPC error rates.

Background

prom-client is the de‑facto Prometheus client for Node.js; Grafana Cloud ships an out‑of‑the‑box Node.js dashboard JSON.

Proposed Solution

1. Add /metrics route exposing default prom-client registry + custom histograms:
   • http_request_duration_seconds
   • rpc_errors_total
2. Docker‑Compose update: Prometheus + Grafana services.
3. Import Grafana dashboard ID 11159; include JSON in ops/.

Acceptance Criteria

• curl /metrics returns ≥ 10 series.
• Grafana panel shows <200 ms p95 latency under load.
• CI job fails if /metrics route breaks.

Labels

chore observability