Implement zero-downtime deployment fallback for edge nodes

When edge nodes experience partial network partitioning during a rolling deployment, the current orchestrator fails over incorrectly, causing ~2-3s of request drops.

Objective: Implement a local cache + async sync fallback that allows edge nodes to serve stale-but-valid responses during deployment transitions, ensuring <50ms p99 latency even during network blips.

Acceptance Criteria

• Edge nodes cache last-known-good state for max 30s
• Automatic fallback triggers on consecutive 5xx responses from control plane
• Metrics dashboard exposes edge_fallback_triggered and cache_miss_rate
• Load tested with 10k RPS under synthetic network partition

Technical Notes

if (controlPlane.healthCheck() === FAIL) {
  edgeNode.enableFallbackMode();
  metrics.increment('fallback_activations');
  logger.warn("Switching to local cache mode");
}

Refer to architecture doc: /docs/arch/edge-resilience-v2.md