Refactor GP Surrogates

[AdrianSosic]

This PR refactors the GP surrogates into their own subpackage, thereby significantly enhancing configurability:

• It introduces the new KernelFactory protocol that is now used by the surrogate and enables flexible context-specific construction of kernels.
• It adds the long-awaited "preset" mechanism that allows to easily switch between a predefined set of GP configurations.
• Based on the preset mechanism, the existing kernel logic is outsourced into corresponding default factories.
• Convenience converters and wrappers are introduced to enable full serializability of the preconfigured settings and use with regular kernels (which is also serializable).

Not included, needs to be done next:

• Adding corresponding LikelihoodFactory and MeanFactory mechanisms to give flexible control of the likelihood and mean models
• Adding at least two more presets: "BOTORCH" and "EDBO" (for the unmodified EDBO setting)
• A new example that creates plots comparing the presets

[AdrianSosic]

Hi @AVHopp, @Scienfitz: I'd say the PR is 95% ready but currently still waiting on the final fixes in #221. But if you like, you can already have a sneak peek and provide some feedback upfront 🙃

[AVHopp]

First round of comments. Have not looked at everything in detail yet.