feat: add Prometheus metrics monitoring and Grafana dashboard to registry server

[yuranich]

What problem does this PR solve?

The registry server had no operational visibility beyond pino logs and PM2 process stats. Operators could not answer questions about model availability, replication progress, per-core contiguous length vs total length, peer counts, full-replica (seeder) counts, RPC request/error rates, or total blob storage without SSH-ing into each node and running ad-hoc scripts. There was also no checked-in PM2 ecosystem config, no reference Grafana dashboard, and no standard way to scrape the registry process.

This branch implements the full monitoring stack pitched in QVAC-17131 and merges it back to main after staging verification.

How does it solve it?

Four layers, all scoped to packages/qvac-lib-registry-server/:

• In-process Prometheus /metrics endpoint on a dedicated HTTP server (default port 9210, bound to 127.0.0.1, configurable via --metrics-host / --metrics-port; 9210 avoids clashing with Prometheus's own 9090 and sits next to pm2-prometheus-exporter on 9209). Wires the Holepunch baseline (hypercore-stats, hyperswarm-stats) for aggregate swarm/DHT stats and adds QVAC-specific instrumentation for per-core visibility (hypermetrics is not wired — it's abandoned upstream and incompatible with Hypercore v11, so per-core gauges are exposed directly):
  • view-core health: registry_view_core_length, registry_view_core_contiguous_length, registry_totals_refresh_age_seconds
  • blob-core replication: registry_blob_core_length, registry_blob_core_contiguous_length, registry_blob_core_peers, registry_blob_core_seeders (peers advertising full replicas), registry_blob_core_bytes
  • catalog + RPC: registry_model_count, registry_rpc_requests_total, registry_rpc_errors_total
  • Blob core is eagerly opened on indexers so length/contiguous_length are populated before traffic arrives.
• External hyper-health-check sidecar as a runtime dependency for independent core discoverability validation.
• PM2 ecosystem.config.js checked into the repo, replacing the ad-hoc per-server setups described in the deployment guide. Defaults are loopback-safe for public-OSS reasons.
• Grafana dashboard (docs/grafana/REGISTRY_DASHBOARD.json) covering the Holepunch baseline plus QVAC panels — view-core replication, blob-core length/contiguous/gap, seeders, peers, bytes, RPC request/error rates — with View Core Replication and Blob Core Bytes promoted to the top of the QVAC section as the first signals operators should read. Deployment guide updated with scrape config and dashboard import instructions.

Net diff against main: 12 files changed, +4161 / −103.

Verification

Deployed to the staging cluster (registry-stg-01/02/03); Grafana dashboard is live and reporting the expected signals. Screenshot from staging attached below.

Test suites on the feature branch:

• Registry-server unit tests: pass
• Registry-server integration tests, including 7 new metrics.integration.test.js cases: pass

Breaking changes

None

[github-actions]

Tier-based Approval Status

**PR Tier:** TIER1

**Current Status:** ✅ APPROVED

**Requirements:**
- 1 Team Member approval ❌ (0/1)
- 1 Team Lead OR Management approval ✅ (2/1)

**Bypass rule:** Triggered (2+ Team Lead approvals (Tier 1 exception)). This PR is approved regardless of tier.

---
*This comment is automatically updated when reviews change.*

[yuranich]

/review