Balance across ionmodes

[niekdejonge]

Many of the changes are pure refactoring, which in the end made it very easy to add the balancing across ionmodes. These refactoring steps are:

• Split the different datagenerators to different files, before they were all in data_generators.py
• renamed SpectrumPairGenerator -> TrainingBatchGenerator, this better captures what the class does.
• Moved the data augmentation to a separate file out of the TrainingBatchGenerator.
• Refactored the data augmentation to make it a bit more modular and testable.
• Moved the Spectrum picking from TraininBatchGenerator into InchikeyPairGenerator and renamed InchikeyPairGenerator to SpectrumPairGenerator.
• Turned the new SpectrumPairGenerator (InchikeyPairGenerator before) into a real generator, before we had a method returning a generator.

After these changes (which didn't add any new functionality), I added the cross-ionization mode capabilities. These can all be found in the inchikey_pair_selection_cross_ionmode.py. This works as follows: 3 SpectrumPairGenerators are made. 1 for pos-pos, 1 for pos-neg and 1 for neg-neg. For the pos-pos and neg-neg this is the same as before, for the across ionmode I had to reimplement the inchikey pair generator and SpectrumPairGenerator. (all in inchikey_pair_selection_cross_ionmode.py). These 3 Generators are combined into a single Generator in CombinedSpectrumGenerator, which just loops over the 3 generators one by one. This CombinedSpectrumGenerator is passed to TrainingBatchGenerator, so TrainingBatchGenerator did not have to be adapted for the new cross ion mode capability.

Note: The new version is fully backwards compatible; The old models work with this version and the new models work on version 2.5.4. So save to merge in that sense.