Add generative models and datasets

[josephbowles]

Major update to the repo to include new generative models and datasets.

New Models

• RestrictedBoltzmannMachine: A wrapped version of sklearn's BernoulliRBM model
• DeepEBM: an energy based model in JAX that uses a MLP network as its energy function and is trained with k-contrastive divergence.

Both models can be scored using the maximum mean discrepancy distance.

New dataset generating functions

These are designed for generative learning

• generate_ising: sample data from an ising distribution via Metropolis Hastings MCMC
• generate_spin_blobs: a binary version of 'gaussian blobs' where configurations are sampled close in Hamming weight to a set of peak configuration

Backwards compatibility
All code from the previous version of the repo should work identically. Training and hyperparameter optimization scripts have been updated accordingly to handle generative models. As a general rule, we specify the model is generative by using y=None in place of labels.

future changes
The repo does not contain any quantum generative models at the moment; it would be good to add a simple example. There is also no changes to the folder structure, so e.g. classifier and generative models are stored in the same directory. It may make sense to change this structure to account for the classifier/generative dichotomy going forward.

[mariaschuld]

Only minor comments, but overall good to go! Thanks for making the changes, and nice to see that it wasn't overly painful to extend the scope of the repo...