Add production-grade observability for Cerebro requests and tool execution

[yacosta738]

Summary

Cerebro has basic tracing logs, but it lacks the observability surface needed for reliable production operations.

Evidence

Current code shows:

• startup/shutdown logs
• some fallback warnings
• TUI event instrumentation

But there is no clear support for:

• request correlation IDs
• structured per-request access logs
• latency/error metrics by endpoint or tool
• Prometheus/OpenTelemetry style export hooks

Expected outcome

• Each request/tool execution can be correlated in logs
• Operators can observe error rates and latency distributions
• The service exposes or integrates with a metrics/telemetry path suitable for production
• Observability expectations are documented for deployments

Suggested scope

• Define a minimal observability baseline for Cerebro
• Add request-scoped structured logging and correlation
• Add metrics or telemetry hooks for request volume, latency, and failures
• Document how to consume that telemetry in production

[linear]

DALLAY-325