Restore owned checkpoint-marker cleanup on handled failures and use MD5 completion markers for recovery.
Release only the acquiring shell’s ckp_running.txt marker in die(), including Bash 3.2 subshell ownership checks. Remove the separate full checkpoint verification pass from load_checkpoint and recreate_db_checkpoint, retain readable sidecar/source checks, and fix gzip sidecar path selection. Update regression tests and operational documentation.
Validation: combined acceptance gate passed; 44 platform stages, 30 topology scenarios, and 112 historical release scenarios across 14 RHEL 8/9 cells passed. All 8 Linux offline suites passed 526 tests each. Focused recovery tests passed on Bash 3.2 and 5.3.