Investigate flaky test "pods should successfully create sandboxes by adding pod to network" in CNI.

There was a PR to try to fix this upstream in k8s, because it happens when creating/deleting pods too quickly, but it was pointed out that this may be a CNI issue instead.

An example: https://prow.ci.openshift.org/view/gs/test-platform-results/logs/periodic-ci-openshift-microshift-release-4.16-e2e-aws-ovn-ocp-conformance/1805919158500593664