feat: dashboard UX optimization + real-time backend stats + health probe fix + pool shuffle

- dashboard.html: major UX overhaul (+657/-308 lines)
- server.py: /api/admin/backends now returns real-time RPM and model_count
- pool_manager.py: random.shuffle backends for load distribution
- config.py: health probe endpoint /v1/models → /models
- docker-compose.yml: add SIDECAR_PRIMARY_WAIT_MAX_RETRIES=6

BIZ-52 post-review optimizations

config.py

@@ -39,7 +39,7 @@ class Config:
     # Health check
     health_check_interval_seconds: int = 60
     health_check_timeout_seconds: int = 10
-    health_probe_endpoint: str = "/v1/models"
+    health_probe_endpoint: str = "/models"
 
     # Admin auth
     admin_token: str = ""