GP hyper-parameter optimization for BMC

[mmahsereci]

Is your feature request related to a problem? Please describe. The current Bayesian Monte Carlo (BMC) implementation does not support hyper-parameter optimization. Thus, it is of little practical value right now.

Describe the solution you'd like. It would be great if the GP-model could fit it's kernel hyper-parameters e.g. by ML type II.

Additional context I suppose it' not clear yet how gradients are computed in ProbNum which are needed to optimize the marginal likelihood to obtain the optimal hyper-parameters. I suppose that is a blocker for this Issue. So mainly opening this Issue for visibility in case someone uses BMC already.