This pull request introduces significant new functionality to the haloflow package, specifically adding a Domain Adversarial Neural Network (DANN) model for domain adaptation tasks. The changes include the addition of new classes and methods for data loading, model definition, training, evaluation, and visualization.

New Functionality:

Data Loading:

• src/haloflow/dann/data_loader.py: Added SimulationDataset class to handle loading and preprocessing of simulation data, and to provide DataLoaders for training and testing.

Model Definition:

• src/haloflow/dann/model.py: Added DANN class for the Domain Adversarial Neural Network model, including feature extractor, label predictor, and domain classifier components.

Training:

• src/haloflow/dann/train.py: Added train_dann function to train the DANN model, incorporating both task-specific and domain classification losses.

Evaluation:

• src/haloflow/dann/evalutate.py: Added evaluate_regression and domain_accuracy functions to evaluate the model's regression performance and domain classification accuracy, respectively.

Utility:

• src/haloflow/dann/utils.py: Added GradientReversal and GradientReversalLayer classes to implement gradient reversal for adversarial training.

Visualization:

• src/haloflow/dann/visualise.py: Added functions to visualize features using t-SNE, including visualize_features_fast and plot_combined_tsne.