Provide domain of kernel parameters

[theogf]

This is an issue to generalize #20 but in a less restrictive way.
I still think that we should provide an hyper-parameter type(s) for the kernels to give informations about the domain.
As pointed out in #20 the transformations given the domain where automatically given in MLKernels.jl and I agree that which should not impose a certain transformation.
However I think it would be beneficial to give the domain of the parameter(s), so one can define these transformations themselves and be able to automatize kernel parameter training.
What do you think @willtebbutt @devmotion ?

[devmotion]

Hmm, I'm not sure what's the best way to approach this though. I can see that it could be useful to retrieve the domain of the parameters, e.g., if one wants to perform optimization in R^n. I thought that TransformVariables.jl might provide some interface for specifying the domain of a parameter/vector of parameters instead of directly defining a specific transformation. That doesn't seem to be the case, it could only be used to specify a set of default transformations for each parameter. This is unfortunate since sometimes one might want to use alternative transformations or even perform constrained optimization.