Huge improvements in python API

Add a ModelBase class to implement quickly a deep net, that will work out-of-the-box in TensorflowModelServe application. The model inherits from keras model class, and can be used as such. Provides a very intuitive framework to code:
- model inputs preprocessing (optional)
- model implementation (mandatory!)
- mode output post-processing (optional)
Harmonize Dataset.dataset and TFRecords classes
- dataset preprocessing function (optional)
- dataset splitting (inputs, targets) to be ready to use in keras models
Some documentation (in the examples/tenso ‎rflow_v2x/fcnn code comments and README)
- How to create a deep net from scratch
- How to train the deep net from patches-images
- How to train the deep net from TFRecords files
- How to convert patches-images into TFRecords files
The API is tested in the CI using the T&F CRC BOOK data

Edited Jul 27, 2022 by Cresson Remi

Resolve "Add a Model class in OTBTF"