Agent registry health check endpoint + auto deregister stale agents

[clintjeff2]

Motivation

• Registered agents were never removed and stale entries accumulated when agents crashed or stopped heartbeating.
• Agents need to signal liveliness via periodic heartbeats and the registry must mark them offline or remove them when they become stale.
• A simple health endpoint and a status=online filter are useful for monitoring and UI filtering.

Description

• Add lastSeen to agent manifests, extend AgentStatus with online, and introduce AGENT_OFFLINE_AFTER_MS (120s) and AGENT_STALE_AFTER_MS (600s) constants in lib/agent-registry.ts and lib/types.ts.
• Implement sweepAgentRegistry() which runs at request time to mark agents offline after >120s and deregister agents unseen >600s, and wire it into listRegisteredAgents, listCapabilities, and getRegisteredAgent.
• Add recordAgentHeartbeat() and a POST /api/registry/[id]/heartbeat route to update lastSeen and set status to online, and add GET /api/registry/health which returns { online, offline, stale_removed_last_run }.
• Add status query filtering to /api/registry and set lastSeen when registering an agent; add tests covering heartbeat, offline marking, stale removal, status filtering, and health counts in __tests__/api/registry/registry.test.ts.

Testing

• Ran npm test -- __tests__/api/registry/registry.test.ts which passed (10 tests).
• Ran the full test runner npm test which exposed unrelated environment/dependency issues (missing lru-cache and @wagmi/connectors) causing other suites to fail; the registry-specific tests still pass.
• Ran npx tsc --noEmit which surfaced the same missing dependency/type errors unrelated to the registry changes.

---

Closes #193

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[clintjeff2]

@leocagli, review and merge.