Add Instance resource agent, dashboards, and resource reporting integration

[zhy1658858023]

Motivation

• Provide real-time resource telemetry from Instance hosts and report snapshots to the Proxy control plane for richer scheduling and pool admission decisions.
• Add convenient local monitoring tools (browser dashboard and Tkinter app) and a small reporter to push the agent snapshots into the Proxy control plane.
• Enrich Proxy and Scheduler control planes with aggregated pool-level resource awareness so scheduling can consider CPU/memory/GPU/network states.

Description

• Add a lightweight Rust resource agent (instance/resource_agent/src/main.rs) that samples CPU/memory/network/GPU and exposes HTTP endpoints GET /healthz and GET /v1/resource/snapshot, plus Cargo.toml/Cargo.lock for the agent.
• Add a reporter script instance/resource_agent/proxy_reporter.py that fetches the local agent snapshot and POSTs it to the Proxy control plane at POST /v1/instance/resource_snapshot.
• Add a browser dashboard server (instance/resource_dashboard/dashboard_server.py), a Tkinter desktop app (instance/resource_dashboard/dashboard_app.py), and static frontend assets (static/index.html, static/app.js, static/style.css) for visualizing instance snapshots and controlling a managed agent process.
• Extend the Proxy control-plane and supporting modules: add POST /v1/instance/resource_snapshot and debug endpoints in proxy/resource/p_control_plane.py, store per-instance InstanceResource and compute build_pool_resource_snapshot in proxy/resource/instance_pool.py, and propagate pool_resource in proxy register/heartbeat flows via proxy/sclient/scheduler_client.py.
• Extend the Scheduler control plane and ProxyPool to accept and expose pool_resource on proxy registration/heartbeat and add debug APIs in scheduler/resource/control_plane.py and scheduler/resource/proxy_pool.py to surface reported pool resources.

Testing

• No automated tests were executed for these changes.

---

Codex Task

[zhy1658858023]

I do not recommend merging this PR in its current form.

Main blockers:

1. The branch is stale and diverged from current main.

   • codex/instance is behind_by=101 relative to main and mergeable=false.
   • Please rebase/update from current main and make the diff a clean increment for Issue #107 only.

2. Scope is much larger than Issue #107.

   • Issue #107 asks for Proxy pool-level resource aggregation and Proxy -> Scheduler reporting.
   • This PR also adds a Rust Instance Resource Agent, Instance dashboards, and a standalone reporter. Those are outside the requested scope and appear to duplicate earlier resource-monitoring/dashboard work.
   • Please remove these out-of-scope additions from this PR unless there is a separate issue explicitly asking for them.

3. proxy/resource/p_control_plane.py appears to reintroduce stale liveness behavior.

   • The added instance-list serialization still uses "is_alive": True if not include_dead else None.
   • For include_dead=true, the UI/debug path needs real TTL-derived liveness, not None.

4. The resource snapshot API shape is narrower than the existing/desired reporting path.

   • The reporter sends only {instance_id, snapshot}.
   • Please preserve any existing metadata fields and avoid regressing resource-report observability.

5. The core #107 path should remain lightweight and backward compatible:

   • Proxy builds a compact pool_resource from InstancePool.
   • Proxy sends it in register/heartbeat payloads.
   • Scheduler stores it per Proxy.
   • Scheduler exposes /debug/proxy_pool_resources.
   • No Scheduler routing policy changes, no Proxy selection changes, no KDN/KVCache/IWS/Instance forwarding changes.

Please redo this as a clean branch from latest main, with only the pool-resource aggregation/reporting changes plus minimal docs/validation notes.