EMR cancel job could cause stale session with SigV4a signer

[noCharger]

The introduction of the sigv4a signer brings to light a potential issue where the request signing, being an asynchronous operation, could be interrupted during Spark context stopping. This issue would manifest primarily when an EMR job is intentionally canceled, leading to stale interactive jobs, although it would not affect the current flow otherwise.

For tracking this feature request, see the issues on GitHub: aws-sdk-java-v2 #5162 and aws-crt-java #782. A swift follow-up is planned for Flint 0.4.