UQ benchmarking on ADME public data

[raquelparrondo]

This PR adds the ADME_public/ subfolder under example_usage/ with files, scripts, and notebooks to reproduce the models and results for the ADME public dataset presented in "Uncertainty quantification in molecular machine learning for property predictions under data shifts" by Parrondo-Pizarro et al.

It includes:

• Two YAML files to create Conda environments for running Chemprop and UNIQUE.
• YAML configuration files to run the UNIQUE pipeline for each ADME public endpoint.
• A Python script to run the UNIQUE pipeline across the five public ADME endpoints.
• Three bash scripts to train Chemprop property models, generate predictions from trained models, and compute latent representations.
• A Jupyter notebook to prepare the ADME public dataset for modeling.
• Three Jupyter notebooks to create input data files for running the UNIQUE pipeline for every property model.
• A Jupyter notbook to reproduce the paper's figures for the public data.
• A README.md explaining how to run the complete pipeline and reproduce the results.