[release/1.5 backport] cgroup2: monitor OOMKill instead of OOM to prevent missing container events

[AkihiroSuda]

Cherry-pick (clean)

• cgroup2: monitor OOMKill instead of OOM to prevent missing container events #6323

When running on cgroup2, currently in a 1-container-pod-with-memory-limit configuration, no /tasks/oom events are generated. This is because the pod cgroup and container cgroups both get the same memory.max setting, and the oom events gets counted to the pod cgroup, while containerd monitors container cgroups for oom events. Fix that by monitoring oom_kill instead and reporting that.
oom_kill events are counted both to the pod and container cgroups.

My test case was the following kubernetes manifest:

apiVersion: apps/v1
kind: Deployment
metadata:
  creationTimestamp: null
  labels:
    app: stress
  name: stress
spec:
  replicas: 1
  selector:
    matchLabels:
      app: stress
  strategy: {}
  template:
    metadata:
      creationTimestamp: null
      labels:
        app: stress
    spec:
      containers:
      - image: progrium/stress
        resources:
          limits:
            memory: "256Mi"
        name: stress
        args:
        - "--vm-bytes"
        - "128m"
        - "--vm"
        - "10"
status: {}

Related issue: k3s-io/k3s#4572

[theopenlab-ci]

Build succeeded.

• containerd-build-arm64 : SUCCESS in 4m 22s (non-voting)
• containerd-test-arm64 : RETRY_LIMIT in 22s (non-voting)
• containerd-integration-test-arm64 : FAILURE in 1m 36s (non-voting)

[estesp]

LGTM