Modular emulator interface

[odunbar]

As we are starting to think about new emulator types, the first thing we should do is make CES agnostic of the emulator type.
A good first step would be to replicate the EnsembleKalmanProcesses modularity for different methods.

In EKP:

1. We first create the process struct with a type (e.g. Inversion orSampler) containing method-specific arguments,
2. We create an EnsembleKalmanProcess struct, which takes in, as an argument this struct of type Inversion or Sampler
3. We write methods for the EnsembleKalmanProcess that depend on the type. e.g update_ensemble(process::Inversion) or update_ensemble(process::Sampler)

Idea for CES:

1. Create structs of types GaussianProcess, RandomFeature, FourierNeuralOperator etc. with all emulator specific arguments
2. Create the Emulator struct, taking in as an argument a struct of type GaussianProcess
3. Create methods for the emulator type e.g. fit or predict for the Emulator

In sampling we will also need to deal with emulators which do not produce uncertainty. (one, perhaps overengineered idea would be to have StatisticalEmulator and DeterministicEmulator type structs)

[odunbar]

(should have closed with #120