Model tuning refinement issues

Some open questions (to be elaborated later) related to the "model tuning" episode:

• Metadata: how complete, and when to we want to introduce this?
• Query the model of their hyperparameters vs hardcoding in the lesson codes?
• "Advanced database format" (all hyperparams, all epoch measurements)?
• Similarity / differences of the output folder structure for the Jupyter-based vs HPC-based tuning?