Total memory value can exceed memory limit in containers endpoint in dcos-metrics

We saw aberrant output from dcos-metrics' container output; total memory value exceeded the memory limit, which ought to be impossible.

On closer inspection it appears that the Mesos /containers endpoint can sometimes not supply the [].statistics.mem_total_bytes field on which metrics relies to populate this value. 

It's unclear why metrics supplies a higher number when this occurs, but presumably a value is leaking somewhere.