DOC: updates to GP

[daniel-saunders-phil]

Issue with current documentation:

The GP docs haven't been updated to the current style as described in here. In addition, there are also some small typos and oversights. For example,

• The Student T process claims keyword args will be passed to MvNormal instead of MvStudentT
• gp.Marginal doc string directs readers to methods for .prior, when it really should be directing them to .marginal_likelihood, .predict.

and a few other small issues like this. I want to take on updating the docs in this area.

Idea or request for content:

I did want input on a few things before I get too deep into the changes.

• All the GP base classes claim you can pass in a 2d array to the covariance function. While that is true, if you set up the GP with a 2d array, it seems like none of the other methods for the GP work. They all throw up an error saying the 2d array is not callable. It might be helpful to include some indication of when this is useful or remove it from the docs.
• jitter. The default value is 1e-6. The doc string don't explicitly indicate what the default is or that the argument is optional. Should we list the default value of 1e-6 on each docstring or should we say JITTER_DEFAULT or is there some other way of indicating the default?
• In gp.Marginal methods, it clarifies that when X is one-dimensional, it really needs to be a (n, 1) column vector. I imagine we should tell users the same thing about gp.Latent methods and the other classes. I just want to check if there is any reason to keep them separate?

I should have a PR ready in the next couple days so please do jump in.