Track intermittent empty sha256sum output in virt/KDM incremental CBT E2E

[coderabbitai]

Summary

The 4.22-virt-kdm-e2e-test-aws job failed while checking payload B after a successful incremental CBT backup. ChecksumBlockDeviceRegion received empty stdout from the read-only dd ... | sha256sum pod exec and returned sha256sum produced no output for /dev/volume0 region in alpine-guestagent/virt-launcher-alpine-guestagent-z5796. The same region was checksummed successfully before the backup, and payload A was checksummed successfully immediately before this failure.

This tracks an intermittent E2E checksum-exec failure, not a confirmed backup data-integrity defect or a confirmed API-server stream defect. The reported PR changes only BSL region handling.

Reproduction evidence

• Job: pull-ci-openshift-oadp-operator-oadp-1.6-4.22-virt-kdm-e2e-test-aws; build ID 2102783124353060864; OCP 4.22.0-0.nightly-2026-09-22-003645.
• Failed test: restore from an incremental (not full) kubevirt-datamover CBT backup in tests/e2e/virt_backup_restore_suite_test.go (Line 1841).
• tests/e2e/lib/virt_helpers.go: ChecksumBlockDeviceRegion rejects empty stdout; execShellCommandInPodWithRetry returns immediately when exec returns no error. Its current retry predicate checks for dial tcp errors only.
• PR: [oadp-1.6] OADP-6065: auto-detect AWS region for CloudStorage-backed BSLs #2461
• Original failure analysis: [oadp-1.6] OADP-6065: auto-detect AWS region for CloudStorage-backed BSLs #2461 (comment)
• Reported by: @kaovilai

Follow-up

1. Check the CI artifacts and reruns for recurrence. Capture stdout, stderr, exec errors, and pod status to distinguish a missing exec stream from a command failure.
2. If an empty successful checksum exec is transient, add a bounded retry for this read-only checksum and a focused test. Keep real command failures visible.
3. Verify that valid checksums still pass and that persistent empty output still fails with useful diagnostics.

The separately reported kubevirt/kubevirt#18957 SourceLost race is not the checksum failure tracked here. Related E2E exec work: #2451.