Scan whole /sys/fs/cgroup directory in the cgroup decoder and use a faster filepath.WalkDir function

[zuzzas]

1. Scan whole /sys/fs/cgroup directory in the cgroup decoder.
   This resolves a problem when you are using a Hybrid systemd cgroup mode.
   Reference: https://systemd.io/CGROUP_DELEGATION/
2. Use a newer filepath.WalkDir function to traverse the cgroup hierarchy. It provides a major speedup to the refresh process.

[zuzzas]

@bobrik

[bwplotka]

I think this will only work if users use cgroups v1. For v2 cgroups we need to look in unified no?

Ignore me unified is child of the dir we look on, in this PR (:

[bwplotka]

Context: I have issues with latency of this endpoint - it takes a huge amount of time to traverse through that refresh when new cgroup id appears. We need faster mechanism here.

[bwplotka]

Ideally with this: bpftrace/bpftrace#1500

[bwplotka]

Added issue: #109

[zuzzas]

@bwplotka
Thanks for the comment. There is a faster implementation of Walk if you are not concerned with Stat()ing every single file. I see a major speedup on a Linux laptop with the newest commit.

root@b5dace1effc1:/src/decoder# go test -run=XXX -bench=. .
goos: linux
goarch: amd64
pkg: github.com/cloudflare/ebpf_exporter/decoder
cpu: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
BenchmarkCgroupRefresh-8              55          19956497 ns/op
PASS
ok      github.com/cloudflare/ebpf_exporter/decoder     1.142s

root@b5dace1effc1:/src/decoder# go test -run=XXX -bench=. .
goos: linux
goarch: amd64
pkg: github.com/cloudflare/ebpf_exporter/decoder
cpu: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
BenchmarkCgroupRefresh-8              24          47062521 ns/op
PASS
ok      github.com/cloudflare/ebpf_exporter/decoder     1.838s

[bobrik]

LGTM. @bwplotka, are you happy with this?