More Stationary GPs

[willtebbutt]

There are a number of kernels in Arno's textbook that we have yet to implement, found in-and-around p260.

Specifically:

• Matern for half-integer powers greater than 5 / 2. i.e. 7 / 2, 9 / 2, etc
• Squared exponential. This is approximated via a high-order Matern, so it's probably a good idea to implement the higher-order Matern kernels first.
• Rational Quadratic. These are approximated in terms of Squared exponential kernels.
• Periodic. They only consider the squared exponential periodic kernel in the textbook -- I wonder if there's a was to do it for arbitrary kernels, in the same way that it can be done usually by mapping the domain onto the circle?
• Noise

A good reference PR is #58 (thanks to @andreaskoher for that)