Release: 0.5.0

• A validator resamples only the model's features when it applies the
  analysis's samplers to each fold, and gives each copied row the other
  columns that the model needs (such as the fixed effects and cluster of a
  fixest model) from the row it copies, as a sampler does for the groups.
  Before, smote and the other samplers that interpolate numbers failed in
  a validator of a fixest model whose fixed effects were text.
• shap_importance (and the SHAP plots, which use it) explains a model that
  needs columns besides its features, such as a fixest model, by
  explaining each row with those columns held at its own values, so its
  SHAP values are the effects of its features given its fixed effects.
  Before, it failed because the explained rows had no fixed effects.
• feature_importance (and importance_plot) leaves out the features that
  a model dropped, such as the collinear features that pyfixest drops,
  rather than failing because there were fewer coefficients than features.