Supply a data.frame or sparse matrix as predictor

[Martin-Jung]

When creating a model, predictors can be added via add_predictors . Currently this function supports only SpatRaster (and eventually stars) objects, but a case can be made that for very large predictors (talking several GBs), it is more efficient to directly supply the extracted covariates (which could also be prepared in any other way).

• Reason is that internally predictors are extracted and saved as data.frame for projection, which however is inefficient for large Raster layers.

• The alternative would be to supply an already extracted data.frame with predictors.

• Predictions in that case will be conducted only with inference_only=TRUE setting in train(). Spatial predictions can then be done via scenario() |> add_predictors() |> project().

• This way also tile predictors could be supplied, thus allowing for better parallization

• Allow data.frames to be supplied to add_predictors

• Check that projections and training work

• Add unit test