Migrate to TF2.x API

[remicres]

I don't have the time right now to do that, but it should not be complicated. External contributions are welcome to develop this feature!

Benefits

• Code easier to maintain
• Less code, simpler code thanks to OTBTF's ModelBase class (Built on top of keras.model)
• Training will benefit from keras framework, hence
  • Validation after each epoch,
  • Model saving strategies (e.g. save best model depending on one validation metric),
  • Training strategies (decay, early stopping, etc)
  • etc

TODOs

• Refactor the model to work using TensorFlow V2 API and OTBTF: the new model must inherit from ModelBase from OTBTF
  • An example how to implement a FCNN model using OTBTF ModelBase here
  • An example how to implement GAN in tf.keras.model here