Create training dataset functionality is too confusing. Hide it.

[Abe404]

Users should not have to think about how to create some kind of intermediary training dataset. They should just have to specify the dataset they want to eventually process (original source dataset) and any temporary or intermediary files should be automatically generated, without the user having to think about or manage these datasets. These could be hidden (or at least not part of standard user workflow) files in a project (for example).

See related discussion #114

[Abe404]

This is a drastic change to the UI and workflow. May take some time.