Experiment with skopt -> sklearn optimizers

Currently the GPR uses [`skopt.Optimizer.tell`](https://scikit-optimize.github.io/stable/modules/generated/skopt.Optimizer.html#skopt.Optimizer.tell) which seems to recomupute the entire fit every time it is called rather than using the present-parameters and adjusting the fit to suit.

This is becoming problematic at later stages of optimisation when there are many steps to fit. There may be methods available in `sklearn` devoted to active learning which we might adopt.

Another option might be [`botorch`](https://botorch.org/)

This issue is to document the need to experiment and explore rather than to implement in the first instance.