cgroup v1 metrics waste memory on excessive calls to isCgroupNSPrivate

[swiatekm]

I was investigating memory consumption of the metricbeat system metrics in Kubernetes, and saw the following heap profile in a cluster using cgroup v1:

These calls happen in a loop here, and look like they can be safely moved out of it. In fact, this call checks the running binary's cgroup, so we should be able to do it once per Fetch, rather than once per target PID.
profile.pb.gz