Add observability to agents server

## Summary
Add baseline observability to the agents server so we can safely run the hosted/cloud deployment for the 0.1 launch.

## Context
From the Agents check-in: the main remaining cloud hardening work is observability: error tracking, log collection, metrics, CPU/resource monitoring, and enough visibility to know whether the agents server is healthy or overloaded.

## Scope
- Add error tracking, e.g. Sentry or equivalent, for agents server failures.
- Add/confirm structured logging and log collection for hosted deployments.
- Emit useful runtime/service metrics, at minimum:
- CPU and memory usage
- request / long-poll load
- durable-stream serving bottlenecks
- agent loop failures
- server/runtime errors
- Add or document dashboards/queries for basic operational health.
- Document known single-node limitations and what we should watch after launch.

## Acceptance criteria
- We can answer “is the agents server running and healthy?” from logs/metrics.
- Exceptions/failures are captured with enough context to debug.
- Basic resource/load signals are visible for hosted agents server deployments.