Adding the Log transformation approach from Osborne et al. 2012

[theogf]

Base on the idea that we should approximate log(integrand(x)) instead of integrand(x)
See : Osborne, M. et al. Active Learning of Model Evidence Using Bayesian Quadrature. in Advances in Neural Information Processing Systems 2012.

Also fixed the issue with the variance which was given as a standard deviation

[johannesgiersdorf]

@theogf I could add the tests later if you find they make sense (atol=1e-1 is not so precise). Otherwise I think we can merge.

[theogf]

Actually I am very uncertain about this implementation...
The paper has a lot of missing details and I think I might overlooked some of them. A group of students was supposed to work on the paper so I wanted to wait for their code to see if they did the same thing as I did