codex/fix docker volume removal race

[drew]

Summary

Retry Docker's transient volume-in-use response during sandbox deletion.

Related Issue

Stability investigation stacked on #3603 and ultimately #3583. The Docker Python E2E job in run 35851188347 exposed a 409 while deleting a sandbox.

Changes

• Retry removal of driver-owned runtime volumes for up to twelve attempts when Docker reports that a removed container still has the volume mounted.
• Continue to fail promptly for other Docker errors and persistent volume conflicts.
• Document the deletion behavior and add a focused classification test.

Testing

• cargo test -p openshell-driver-docker volume_removal_retries_only_mount_release_conflicts --lib: passed.
• mise run pre-commit: passed.
• mise run ci: passed.
• Focused local Docker Python E2E test_sandbox_labels_and_selectors: passed.

Checklist

• Signed-off conventional commit
• Documentation updated
• Full local CI passed
• Focused Docker E2E passed

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[drew]

Superseded by clean stacked PR #3621 in GitHub Stack #3623. The selected fix is preserved there without the intervening campaign changes.