Dimensions for y variable in gp.marginal_likelihood

Currently no… but really trying to get this in the next major release. In the meantime this might be helpful.