ateom-microvm: "waitid: no child processes" flakes e2e restore

[HavenXia]

Expected Behavior

E2E tests (micro-VM) passed.

Actual Behavior

E2E tests (micro-VM) is flaky, example failure.

Steps to Reproduce the Problem

1. This flake shows up after some pr merged.

Symptom

e2e-test (mtls) flakes on main in the demo suite's TestActorLifecycle restore
subtests. The ResumeActor RPC fails with:

Internal: while staging overlay lower for "counter": bind-mounting image rootfs
".../bundles/counter/rootfs" -> "/run/kata-containers/shared/sandboxes//shared/counter/rootfs":
waitid: no child processes ()

[HavenXia]

I don't fully understand where the problem happens but the different I saw is we have reapLock.RLock() in gvisor implementation but it's missing in microVM implementation.

In ateom-microvm/main.go, there's comment that

// reapLock guards subprocess exec against the child reaper: ateom-microvm
// spawns the cloud-hypervisor process under it.
reapLock sync.RWMutex

[aramase]

Haven Xia (@HavenXia) I opened a PR, PTAL!

cc Benjamin Elder (@BenTheElder) for the changes in ateom-microvm.