Implements #60 authsession garbage collection. If we assume that an authsession is around 1024 bytes (this assumes a 16 char name + groups + claims) then this means that in 1Gb of ram we can store about 1 million in progress auth attempts. Obviously, we don't want infinite memory growth, but we can't use an LRU cache due to the future desire to use concurrent trees. So instead we prune the tree based on a timeout when we start and auth operation. Auth session id's are generated from a timestamp similar to how we'll generate replication csn's. We can then apply a diff that will split all items lower than the csn/sid and remove them from future consideration.
We set the default timeout to 5 minutes. This means that assuming 10,000 auths per second, we would require 3GB of ram to process these sessions before they are expired. We expect any deployment with such large loadings can affort 3Gb of ram :)
This adds support for authentication and credential storage to the server.
It also adds account recovery and options for integration test fixtures, refactors
to make the client library easier to manage, and support clean seperation of the
proto vs lib.
