Making it possible to push metrics in batches rather than one at a time

[wk8]

When using dogstatde to push a lot of custom Datadog metrics, we've
observed that the overhead of checking workers out & back in of the
dogstatsd_worker wpool can become a performance bottleneck.

This patch makes it possible to submit several metrics of the same type
at one time, hence allowing for a single worker check-out/in cycle to submit
them all, for example:

dogstatsd:gauge([{"users", UserTypeCount, #{ user_type => UserType }}
                 || {UserTypeCount, UserType} <- UserCounts]).

We've deployed that in production, and found a noticeable difference in our
use case.

This patch should be entirely backward-compatible.

Updated tests, and added a few more.

[wk8]

Hey @waisbrot long time no see ;)

[JoshRagem]

this looks great, thanks!

[wk8]

@JoshRagem : thanks for quick merge, but found a stupid bug => #24 ... sorry!!

[waisbrot]

@wk8 can you quantify "a lot"? I'm thinking that if it's far more than most people would be sending then it'd be nice to update the README so that users have an idea of when they need to worry about batching. And if it's a fairly low number we could look for ways to make batching happen automatically.

[wk8]

@waisbrot we're seeing wpool starting to be a bottleneck at about 2 or 3 thousand metrics per minute, FWIW

[wk8]

@waisbrot if you want => #25

[wk8]

As to "look for ways to make batching happen automatically" => our own dogstatsd wrapper does just that, a simple gen_server that we push metrics to, and which then pushes those in batches to dogstatsd, either each 5 seconds or when the list grows bigger than 200, whichever happens first. Happy to add that to this repo, just wasn't sure if you guys would want that.