executor: detect cgroup leaks

[dvyukov]

It's very easy to forget to uncharge a memory cgroup, as the result it will leaks and lead to memory leak.
It's hard/impossible to catch missing uncharges red handed. Instead we could try to check total number of cgroups in /proc/cgroups and catch leaked cgroups.
Unclear if it's feasible to do a normal non-snapshot mode. We would need to stop all test processes, wait for them, do full cleanup (including removing root cgroups), and then do the checking.
Maybe it's easier to do from time to time (e.g. during smash) in snapshot mode only.

Other cgroup types can be checked in a similar way.
Question: are there other resource types that can be checked for leaks in a similar way? (namely: in shapshot mode: kill/wait test process, do all possible cleanup, check that resource usage drops to the before-the-test level).