Split the package installation

In #108, we added a new directory ml_models to store experimental nn.Modules. This was done to support swapping a physical component, not to train or validate the wofost crop model. The models in ml_models are therefore experimental, and results from the wofost crop model using them are not validated in this package. We keep ml_models in the main branch to maintain them and include them in releases, Zenodo publications, and documentation. It also serves as a reference for researchers on how to develop and integrate their own models into the diffWOFOST ecosystem.

If ml_models grows, it could make installation heavier and harder to maintain quality (e.g., testing on GitHub runners). One option is to split it into extras, for example:

pip install diffwofost (physical models only)
pip install diffwofost[ml_models] (physical + ML models)

This isn’t urgent and can be addressed later.

Start by locating the package metadata and current installation and release configuration, then inspect how ml_models is included and tested on GitHub runners. Compare the base and ml_models installations, and verify that releases, Zenodo publications, and documentation still include the experimental models appropriately.