excessive metrics sent to statsdaemon, leads to high statsdaemon usage

[Dieterbe]

(this is using the carbon input plugin for ingestion)
https://snapshot.raintank.io/dashboard/snapshot/aqlU18OPIdtyaZ6vtMOYZZGnNERZfx6B
=> large amounts of timers to statsdaemon. those cause the vast majority of cpu since they require most computation.

with ngrep -d any -W byline port 8125 | grep 'ms' | cut -d ':' -f1 I see almost all timers are either:

1. carbon.metrics_per_message (can be removed since it's always 1)
2. idx.memory.get_duration
3. idx.{memory,cassandra}.add_duration

solving this should significantly reduce statsdaemon cpu usage.
we can also look at some other metrics, like some counts, though i suspect those will make less of a difference.
anyway, can easily be done in raintank-docker.

[Dieterbe]

this requires adding rate args to the statsd functions and their invocations. i'm going to wait for #320 since that is such a huge changeset, otherwise we'll have a nasty conflict

[Dieterbe]

now that #400 is stabilizing, started working on this. will just get rid of statsd[aemon]. early WIP in #384

[Dieterbe]

fixed via #384