e2e: cozyreport keys satellite files by node, so a second pod on the same node overwrites the first one's output #4743

When two LINSTOR satellite pods run on the same node, cozyreport writes both of their reads into one file and the second pod wins. That happens during a DaemonSet rollout, while the old pod is still Terminating next to the new one.

The satellite loop in hack/cozyreport.sh (main, L2097-L2117) names each pod's output after its .spec.nodeName, as linstor/error-reports/<node>.tgz. cozyreport_read_exec opens its target with >, so the later pod in the list truncates what the earlier one wrote. If the later one is the Terminating pod, its container not found error replaces a good bundle. The comment above the loop says a collision would still land in a distinct file. That holds only for the pod-name fallback, not for two pods that report the same node.

The per-node ZFS listing from #4657, linstor/zfs/<node>.txt, uses the same key, so it has the same problem once that PR merges.

Nothing in the loop notices that a node's file was already written in this run. The fix belongs in the loop, so both files get it at once: skip pods with deletionTimestamp set, or fall back to the pod name when a second pod maps to a node already seen. The lanes don't roll satellites during collection today, so this is unlikely to bite in CI.