fix(k8s): replace blocking time.sleep with asyncio.sleep in _wait_for_sandbox_ready

When the workload is not yet visible after creation, the polling loop
fell into a time.sleep branch instead of await asyncio.sleep. This
blocked the event loop during sandbox creation, preventing FastAPI from
handling other requests (including /health), which could cause liveness
probe failures and pod restarts.

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

Author: qingyuppp

server/opensandbox_server/services/k8s/kubernetes_service.py

@@ -196,7 +196,7 @@ async def _wait_for_sandbox_ready(
 
                 if not workload:
                     logger.debug(f"Workload not found yet for sandbox {sandbox_id}")
-                    time.sleep(poll_interval_seconds)
+                    await asyncio.sleep(poll_interval_seconds)
                     continue
 
                 status_info = _normalize_create_status(