t-SNE plot improvements?

[alanngnet]

Ideas for improving the _cluster_plot() function I added to src/eval_testset.py :

• investigate whether PCA reduction would be valuable, or in which circumstances (maybe label count>N?) it's needed. Add it if so.
• add option to training script to save an epoch-specific cluster plot at each epoch that calls eval_for_map_with_feat
• connect extract_csi_features so _cluster_plot() can have awareness of exactly which labels and which samples were excluded from the training data and mark those in the plot. #13
• write a separate script to use the saved dist_matrix and ref_labels to do more customizable t-SNE plots.