Improved development features for new models

This release include new/improved development feature to make training new models easier.

List of changes:

• Improved function to generate dataset for training/testing new models (defect_detection.generate_dataset)
  - Supports taking images from different sources and combined them into one dataset
  - Improved data augmentation feature

• Improved function to load a batch of images from disk directly to the required device (defect_detection.load_batch)
  - Support option to automatically generate a noisy image batch (useful for training new models)

• Added function to initialize and train a new model on a given dataset (defect_detection.deepAE_train)
  - The structure of the model must be defined in a configuration file
  - The provided training loop supports automatic input noising, learning rate decay and training/validation split