feat(phase-2): workstream 7 — Prometheus + Grafana Monitoring

- Add prom-client 15; shared registry in src/metrics/registry.ts (7 metrics)
- HTTP request counter + duration histogram via metricsMiddleware
- DB query duration histogram wrapping pg Pool.query
- Redis command duration histogram via typed instrumentRedisMethod wrapper
- agentidp_tokens_issued_total in OAuth2Service
- agentidp_agents_registered_total in AgentService
- GET /metrics unauthenticated endpoint (Prometheus text format)
- docker-compose.monitoring.yml overlay (Prometheus + Grafana)
- Grafana auto-provisioned datasource + pre-built AgentIdP dashboard
- docs/devops/operations.md monitoring section added
- 36/36 unit tests passing, 100% coverage on new metrics code
- Fix pre-existing unused import in tests/integration/agents.test.ts

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/services/AgentService.ts

@@ -19,6 +19,7 @@ import {
   AgentAlreadyDecommissionedError,
   FreeTierLimitError,
 } from '../utils/errors.js';
+import { agentsRegisteredTotal } from '../metrics/registry.js';
 
 const FREE_TIER_MAX_AGENTS = 100;
 
@@ -81,6 +82,9 @@ export class AgentService {
       { agentType: agent.agentType, owner: agent.owner },
     );
 
+    // Instrument: count successful agent registrations
+    agentsRegisteredTotal.inc({ deployment_env: data.deploymentEnv });
+
     return agent;
   }