Rafactor(NeuralNetwork): remove workers, batch_queue_size, multiprocessing parameters

[rizoudal]

This pull request removes the parameters workers, batch_queue_size, and multiprocessing from the neural network class and all related files. These parameters are no longer supported in the latest versions of TensorFlow and Keras, and their removal resolves compatibility issues.

Additionally, this pull request refactors the code to replace the usage of the .hdf5 file format with the .keras format across the project.

Please review the changes and let me know if there are any questions or further adjustments needed.

[muellerdo]

1. Classifier buggy? -> input_stack = [input_stack, self.metadata[index_array]] muss in klammern (input_stack, self.metadata[index_array]) in zeile 304 vom datagenerator
2. albumentations ai.pad(aug_image, org_shape[0], org_shape[1]) in image_augmentation, value=0, pad_mode="edge"
3. xai problem: gradcam & co (gradienten basiert), find last conv layer -> class variable output_shape not existent anymore -> how can we find the last convolutional layer in a random architecture? is there an alternative to output_shape class variable to get the shape of a tensor?