Logged in users may be more likely to have client side errors - either
through using interfaces that require authentication or through gadgets
that they run. I'd like to separate the two.
Bucketing like this means we can still continue to count all client
side errors, but we can have additional graph lines for logged in.
This way we can notice spikes in bugs limited to logged in users
for example T216853
Change-Id: I965c45f8b548abb16b400571ddf7852ca088529b