docker stats API caches the stat and reading it frequently gives back the same value.
this trips our "rate" code into thinking rate is 0.
one approach was to persist polling like we do now and ignore entries based on stats.read.
this works fine but the streaming approach seems better since we don't need to poll anymore.