fix(sandbox): reject stale exit during restart

[drew]

Summary

Prevent a delayed canonical-process exit from the stopped sandbox instance from racing with restart and moving the sandbox from Starting to Error.

The Kubernetes v1beta1 combined E2E reproduced the same failure twice on the unpatched commit:

• https://github.com/NVIDIA/OpenShell/actions/runs/32434526234/job/96635981525?pr=2822
• https://github.com/NVIDIA/OpenShell/actions/runs/32434526234/job/96639911918?pr=2822

Related Issue

No issue required: this is a localized lifecycle race with a focused regression test.

Changes

• Preserve the previous main-process instance ID as a tombstone through stop/start transitions
• Ignore a matching exit report from the stopped instance while the sandbox is Starting
• Continue accepting an exit from a different instance so genuine failures of the new process remain terminal
• Add a regression test for both the stale-old and legitimate-new exit paths

Testing

• mise run pre-commit passes
• cargo test -p openshell-server --lib passes (1,406 passed, 8 ignored)
• Unit test added for the restart race
• E2E checks run in GitHub Actions

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)

[github-actions]

Label test:e2e applied for 8fc8528. Open the existing run and click Re-run all jobs to execute with the label set. The run will execute the standard E2E suite after building the required gateway and supervisor images once. The matching required CI gate status on this PR will flip green automatically once the run finishes.