Kubernetes Pod Operator fails with 404 errors when pods are preempted by daemonsets

### Apache Airflow Provider(s) cncf-kubernetes ### Versions of Apache Airflow Providers apache-airflow-providers-cncf-kubernetes>=8.3.2 ### Apache Airflow version 3.x (also affects 2.x) ### Operating System Linux (any Kubernetes distribution) ### Deployment Astronomer ### Deployment details Kubernetes cluster with Karpenter or similar node autoscaling. Airflow running on Kubernetes with KubernetesPodOperator tasks. Nodes are dynamically created and daemonsets (calico-node, aws-node, falco, etc.) are scheduled on new nodes. ### What happened Kubernetes Pod Operator tasks are failing with "Pod Not Found" (404) errors when pods are preempted by higher-priority daemonset pods on newly created nodes. **Error logs:** ``` [2025-12-02, 11:00:52 UTC] {standard_task_runner.py:110} ERROR - Failed to execute job 267126 for task fresh_squeeze ((404) Reason: Not Found HTTP response body: {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"pods \"fresh-squeeze-vqfekgln\" not found","reason":"NotFound","details":{"name":"fresh-squeeze-vqfekgln","kind":"pods"},"code":404} ``` **Root cause:** 1. New Kubernetes node is created (e.g., via Karpenter autoscaling) 2. Kubernetes Pod Operator task pod gets scheduled on the new node immediately 3. Daemonset pods (calico-node, aws-node, falco, etc.) begin scheduling on the same node 4. Daemonsets have higher priority than task pods 5. Task pod gets preempted by daemonsets 6. When the operator tries to read the pod status, it receives a 404 error because the pod no longer exists 7. Task fails even though this is a transient infrastructure issue **Kubernetes events show:** ``` k get events --field-selector reason=Preempted -A NAMESPACE LAST SEEN TYPE REASON OBJECT MESSAGE frigid-comet-7713 4m33s Normal Preempted pod/reltio-y1f5wtrb Preempted by pod 38606146-2e79-4ee3-824c-abb1ea96740b on node ip-10-105-68-155.us-west-2.compute.internal ``` ### What you think should hap…