Internal refactoring - create a dedicated handler for machine level stats

[vishh]

As of now machine level stats are split across multiple packages - raw container handler for / cgroup stats, fs/fs.go for filesystem stats, cpu load from manager package.

By adding a dedicated handler for the machine, we can group all the machine stats logic into a single package. This will make it easier for devs to add new features.

[timstclair]

+1 - This will also let us simplify the container handler interface by removing some root-container specific functions. If I get some time later this week I'll take this on.

/cc @sjpotter - FYI