Logs not uploaded to S3 on failed retries

### Apache Airflow version 3.1.0 ### If "Other Airflow 2 version" selected, which one? _No response_ ### What happened? We are experiencing a critical issue with log upload to S3 when using DbtTaskGroup from Cosmos in our Airflow 3.x environment (Cosmos 1.10.2, dbt-core 1.11.x, Astronomer runtime). The airflow deployment is running on EKS in AWS and tasks in airflow are run as KubernetesExecutor (tried with CeleryExecutors and exact same problem was observed). When a Cosmos DbtTaskGroup task fails and is retried, the logs for the failed attempts are not uploaded to S3 (remote logging enabled). Only the logs from the last retry are available in the Airflow UI and on S3. This does not happen with standard Airflow operators (e.g. BashOperator), only with Cosmos DbtTaskGroup. What is weird is that if we choose to put no retry at all, we have the logs the first time. But if we choose to put one retry on the airflow dag, we lose the first attempt but we get the second. During the execution of the tasks, we have the logs because airflow is still pulling them directly from the Task pod. But as soon as finished, if the pod have an error, it crashes before being table to upload any log to S3. So logs of the attempt are lost. The pod running the task is stopped abruptly after failure, before logs can be uploaded. We see this error in the logs: airflow.sdk.api.client.ServerResponseError: Server returned errormessage': 'Not Found', 'detail': {'detail': 'Not Found'}}...File "/home/airflow/.local/lib/python3.12/site-packages/airflow/sdk/execution_time/comms.py", line 258, in _from_frame Full stack trace of the failing pod not succeeding to upload its logs to S3 : {"timestamp":"2025-10-07T09:37:34.134606Z","level":"warning","event":"The `airflow.models.baseoperator.BaseOperator` attribute is deprecated. Please use `'airflow.sdk.bases.operator.BaseOperator'`.","category":"DeprecatedImportWarning","filename":"supervisor.py","lineno":1739,"logger":"py.warnings"} {"timestamp":"2025-1…