Add basic feature exploration framework

[DavidLandup0]

Adds the FeatureAnalyzer class, which accepts:

• A model (PyTorch or TensorFlow)
• A dataset (tf.data.Dataset or torch.utils.DataLoader)
• Components (Learned features -> projected to 2 or 3 using PCA and t-SNE)
• Batch limit (No need to run through the entire dataset)
• Backend flag

And outputs a 2D or 3D plot of PCA and t-SNE analysis of the learned features, a la:

feature_analysis = deepvision.evaluation.FeatureAnalyzer(pt_model, 
                                                         train_dataloader, 
                                                         components=3,
                                                         limit_batches=1,
                                                         backend='pytorch')

feature_analysis.visualize()

This class is meant to be used with any model that has a top (or that has a top taken off, which tends to work much better for general feature analysis)

TODO: look into options of adding human-readable labels besides targets for legends as well as options of plotting images instead of scatter plot markers