Implement RSC (lightly tested) / Cleanup hyperparameter registry

[SirRob1997]

This PR implements the following features:

• Representation Self-Challenging (RSC) from: https://arxiv.org/abs/2007.02454
• Cleanup algorithm-specific hyperparameters in hyperparameter registry so they don't flood the console by default
• Adapted MNIST_CNN forward pass to use AdaptiveAvgpool2d instead of mean, this allows for more flexible usage in algorithms and feels more clean

You might also want to delete the MLP parts in networks.py and hparams_registry.py as well since I think MLP was only added for Debug datasets that don't get used anymore?

[lopezpaz]

Note: MLP is used in DANN variants.

[SirRob1997]

Then the MLP parameters should only show in the console when DANN is used as an algorithm

[lopezpaz]

Merged, although we will have to find a solution to guarantee that the same random parameters are sampled in hparams_registry given the same random seed, for all future commits. Right now, adding algorithms or conditional logic in hparams_registry changes the random hyperparameters sampled for a given algorithm, dataset, and seed.

cc @igul222

[lopezpaz]

@SirRob1997: Preliminar runs show a very poor performance (39%) at OfficeHome. The paper reports 63%. Could you investigate how to get a higher number in this dataset with default hyper-parameters?