Queue Placement on Storage Nodes

This is essential for scaling and operating a large deployment.

We could use MongoDB's built-in-sharding, or create our own placement algorithm if we need more control since we will likely want to split chunks not based on chunk size, but on request rate.

I propose this be implemented as a placement function/class that is passed into the storage driver on init, and that this placement function/class have it's own configuration section in marconi.conf

This lays the groundwork for supporting other data stores that, unlike Mongo, do not have built-in sharding logic.

Trello: https://trello.com/c/9hl3oizG