Fix machine stats collection

[pjenvey]

With the new Heka+InfluxDB line protocol, the machine stats (from /proc) are no longer captured. We either need to fix heka or investigate something else to collect these (e.g. collectd).

[pjenvey]

cAdvisor looks promising, e.g.:

https://www.brianchristner.io/how-to-setup-docker-monitoring/

[pjenvey]

influx's own telegraf looks even more promising. Supposedly it uses less cpu/mem than cAdvisor.

It appears to not only support capturing of machine stats but can also collect statsd input -- potentially replacing our heka setup entirely.