ML Splits command should be more explicit on errors from backend

Refs https://redmine.teklia.com/issues/7098

The command arkindex elements ml-splits should catch the backend errors with 400 status code, and display them a bit more nicely.

When a non_field_errors with content ['A dataset with this name already exists in this corpus.'] is present, it should also display an extra help text on a new line: "The dataset name "Training dataset" is already used. Please set a different one using --dataset-name"

---

Second, the documentation should state that this parameter is heavily recommended to set, as a dataset name must be unique across a corpus