Final model: save feature names for interpretability

If a final model is produced, it can be loaded like:

```{python}
from drevalpy.models import MODEL_FACTORY
enet_class = MODEL_FACTORY["ElasticNet"]
enet = enet_class.load("path/to/final_model")
enet_model = enet.model
print(enet_model.coef_)
```

Sklearn models have the option of saving feature_names_in_ when X in fit() has str names. In the gene expression-based models, we select the features with the gene list so at least it's possible to reconstruct the features. In the proteomics models and multi-omics models, it's variable.
Same for MOLIR, SuperFELTR which use the VarianceSelector.
Overall, it would probably be nice and reproducible to save the names and order of the input matrix.