Lazy Loading for WflowSBM_HPC returns DataArrays instead of Tensors if lazy loading is active

[lineick]

If lazy loading is true the following conversion does not happen:

        if not self.cfg.data_lazy_load:  # Only if we're not doing lazy loading
            # Convert xarray to pre-processed tensors
            self.xd = torch.from_numpy(
                self.xd.transpose("time", "feat", "lat", "lon").values
            )
            self.y = torch.from_numpy(
                self.y.transpose("time", "feat", "lat", "lon").values
            )
            self.xs = torch.from_numpy(
                self.xs.transpose("feat", "lat", "lon").values
            )

This leads to a xarray.DataArray being returned instead of a tensor in get_item.

It can be easily fixed by converting to tensors on the fly in __get_item__, however this is suboptimal as such a conversion would be triggered for each data sample.