Add embedding evaluation model

[florian-huber]

• new EmbeddingEvaluatorModel (Inception Time CNN)
• new LinearModel
• new MS2DeepScoreEvaluated matchms-style score --> gives "score" and "predicted_absolute_error"
• linting and minor documentation changes

[niekdejonge]

Great work Florian! Mostly some comments about adding documentation. And some potential future changes.

[niekdejonge]

SettingsMS2Deepscore is already stored in ms2ds_model, so we could just retrieve the settings from there, instead of passing it as a separate parameter.

[florian-huber]

That's true. But somehow also a weakness of the on-for-all setting object. For us that might be handy, but I could imagine that this hides a bit of what's happening to new users. What do you think?

[niekdejonge]

Yes, good point, I see how it can be confusing. I do actually not expect many people to dive that deep into the codebase. Some will train new models, but I would expect them to use the wrapper functions that we provide. So to me, the most important people to design for is ourselves in the future. So the design that is clearest to us in 1 or 2 years. Still I do not know which of the two that is... I would use the settings in the model, or I would make a separate evaluatorSettings class. We now save them separately, but still save all the settings of the "other" models in there as well, which is a bit unexpected.