Training workchain to do the following steps:

• geometry optimisation with QE of many files, then output the structures together in a xyz file (?)
• handle errors from the calculations, can the training go on even if some calculations fail? #155
• Use said xyz file to train a model using janus
• output and save the model
• other steps?

External plugins possible problems or things to consider

• geometry optimisation problems handled by aiida QE when possible, so that's good
• that problem with aiida-qe and janus coexistence https://github.com/stfc/janus-core/issues/196
• submission for many jobs could be handled through aiida-hyperqueue (that I think the user sets up indipendently) or just something like aiida submission controller that I have to look into (possibly it's still a user's problem anyway)
• gpu? that's handled in janus I think but never tested how it works?
• tests for submission in a queue (I'll see what other plugins do about that)