Add /count API endpoint

[soaxelbrooke]

While counting is best done in bulk, its pretty clear there are instances where you'd prefer services to be interfacing with phrase over HTTP instead of relying on running a cron job to keep counts up to date.

[soaxelbrooke]

One big question: persistence - we want to make sure that counting is as atomic as possible. But, reading a whole partition and then writing it so you can count the document "love it" would be a waste. Perhaps we can keep a WAL of sorts in a sqlite database?

[soaxelbrooke]

Let's not overdo it for now - we can leave the batching to the client for now, and address it if its brought up later.