Further improvement to DNN/BDT code structure

[andreypz]

I notice that the params/trainings.yaml file is, in fact, not used. Instead the model files are hardcoded in the config.

• So, we should switch to using the yaml file to read all the parameters

• We should understand whether the following Warning is important:

[WARNING ] Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.

• Do the training for all 4 periods of Run3
• Implement a feature to only use events with even Event Number for training. The Odd event numbers should be used for analysis. Then, ta scale factor of 2 should be also implemented for the affected samples.

[ValVau]

Sure, I will take care of that