New command to create train/test/val element splits

Refs https://gitlab.com/teklia/requests/-/issues/809

A common task for ML Scientists is to create a new folder on arkindex with 3 sub-folders (train / test / validation). These folders are then used by training workers to build a new model (this is the poor man dataset management.

We need a script that takes as input:

• either a corpus id or a folder ID that will serve as source for the elements
• optional multiple --element-type, defaults to ["page", ]
• ratios to pick elements and assign them to folders:
  • --train-ratio, defaults to 0.4
  • --test-ratio, defaults to 0.3
  • --validation-ratio, defaults to 0.3
• optional --folder-name, defaults to "Training dataset"
• optional integer --nb-elements, default to None

The command workflow is like this:

• list randomly elements that will be used:
  • from the corpus when set
  • from the folder ID when set
  • by using the element types available
  • until you reach the nb_elements mentionned:
    • when a positive integer is provided
    • no limit when it's set to None (list all elements)
• once you have all the element IDs, split them amongst the 3 destination, by applying the provided ratios
  • ⚠ the sum of ratios must be equal to 1 !
• create the training folder at the root of the corpus using the mentionned name
• create the tree sub folders (Train, Test, Validation)
• finally, link all the elements in their respective folders
• display a full link towards the Training folder