Fix named cgroup handling in docker (#980)
While named cgroups work fine outside of docker container, they weren't
working when created and used inside of a docker container. The problem
was caused by the fact that the hierarchy root extracted from
/proc/self/mountinfo and the cgroup path extracted from /proc/self/cgroup
are not equal for named groups. They just share the same prefix.
The cgroups handling code was not epxecting this case and ended up building
the final cgroup path incorrectly (including the common part of the path).
This change fixes it by checking for matching prefix of the paths instead
of comparing the whole paths.