Improve bulk doc saving, retrieval

[Jason Gerard DeRose]

Although CouchDB doesn't have the greatest serial performance, it's generally been fast enough for Dmedia/Novacut so far. Certainly fast enough for how we use it from the UI, and generally fast enough for how we use it in the backend.

However, there are some recent Dmedia features that require us to retrieve and (possibly) update many docs in rapid succession, and the serial CouchDB performance just isn't good enough for this.

This bug is to track the change, which:

 * Changes Database.bulksave() to use "non-atomic" semantics and raise the new BulkConflict exception when there are conflicts

 * Adds experimental Database.get_many() method to make it easier to retrieve many documents at once