fix: accurate instance status — STARTING until pod is actually running

Instance status now reflects actual container state:
- startOne() sets STARTING (not RUNNING) after container creation
- syncStatus() promotes STARTING→RUNNING when pod is ready
- syncStatus() demotes RUNNING→STARTING if pod restarts (CrashLoop)
- External servers still get RUNNING immediately (no container)

Previously, CrashLooping pods showed as RUNNING in mcpctl get instances.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/mcpd/tests/mcp-server-flow.test.ts

@@ -484,7 +484,7 @@ describe('MCP server full flow', () => {
       expect(instancesRes.statusCode).toBe(200);
       const instances = instancesRes.json<Array<{ id: string; status: string; containerId: string }>>();
       expect(instances).toHaveLength(1);
-      expect(instances[0]!.status).toBe('RUNNING');
+      expect(instances[0]!.status).toBe('STARTING');
       expect(instances[0]!.containerId).toBeTruthy();
 
       // 3. Verify orchestrator was called with correct spec
@@ -564,7 +564,7 @@ describe('MCP server full flow', () => {
       expect(listRes.statusCode).toBe(200);
       const instances = listRes.json<Array<{ id: string; status: string }>>();
       expect(instances).toHaveLength(1);
-      expect(instances[0]!.status).toBe('RUNNING');
+      expect(instances[0]!.status).toBe('STARTING');
       const instanceId = instances[0]!.id;
 
       // Delete instance → triggers reconcile → new instance auto-created