Create a Skin Condition classification task

[JulienVig]

Add a skin condition multi-class classification task as a predefined task in Disco for the upcoming demo. The SCIN dataset is described here.

• original model previously used for this task currently only predicts the benign class
• create a sample image dataset to provide to users (original dataset is 3GB and requires creating an account on Kaggle and creating the csv label mapping file) -> Available on the google cloud bucket
• Testing a model fails with error
• fine tune architecture and hp. I tried using a pre-trained mobilenet (a334219, commented code in skin_mnist.ts) but encountered Discrepancy between training logs and manually evaluating metrics #679 and decided to settle on a smaller handmade CNN for now.
  For 800 images
  
• Allow connecting a folder rather than individual files -> moved to Connecting folder displays alarming "Upload" warning message #680
• Add an example of how to create csv file to connect data -> docs/dataset_csv_creation.ipynb
• At test time, show label names rather than integer numbers
  Before:
  
  Now:
  

[tharvik]

well done (and sorry that you had to rush it)! a bit of cleanup but nothing vital

[JulienVig]

@tharvik do you want to review the latest changes or shall I merge the PR?