Fix the cleanup of network interfaces on pod deletion

When the podagent get called to dynamically attach a network interface to a running pod than the pod get deleted, kactus was getting confused and was not deleting all network interfaces in a pod as it should before the linux network namespace get destroyed. This was caused by the libcni caching the results of the cni-plugin calls under /var/lib/cni/results and loading the cached result of the cni-plugin invoked by the podagent instead of the one related to eth0, use separate cache file to avoid this problem.