Performance optimizations

[mileslucas]

I have mainly been writing this code to get up to speed with some related research work, so I don't have as much time to try and optimize it after it works. The runtime is about equivalent to when I used nestle in PyCall.jl, but I've never looked at type-stability or optimization. The Bounds.Ellipsoid code has quite a few operations, and so does StatsBase.step!.

If someone wanted easy pickings for contribution, this would be a great opportunity!