Linux datapath tests can accept terminating pods from a previous invocation

[Arleee1]

Summary

Consecutive IPv6 and IPv4 datapath tests reuse Goldpinger labels. Asynchronous cleanup can leave terminating pods that satisfy the next invocation's readiness check, followed by an intermittent Less than 2 pods on node failure.

Steps to reproduce

Using the pre-fix code on a dual-stack cluster with at least two Linux nodes in nodepool1, run:

go test -count=1 -timeout 3m -tags=connection,integration -run '^TestDatapathLinux$' ./test/integration/datapath -isDualStack=true
go test -count=1 -timeout 3m -tags=connection,integration -run '^TestDatapathLinux$' ./test/integration/datapath

Keep the same namespace and pod-name settings. The failure depends on old pods still being visible when the next invocation starts; it does not occur every time.

Expected behavior

Cleanup waits for the previous invocation's pods to disappear before returning.

Relevant code

Pre-fix cleanup and pod helpers.

Proposed fix

#4863 adds WaitForPodsDelete after the existing controller deletions, using both workload selectors and t.Error on failure. Both selectors cover custom -podName settings.

Readiness checks, RBAC cleanup, and timeouts are unchanged.