[ENH] Add preprocessing steps (scaling/transformations) to D2 layer

[ravjot07]

This PR implements target normalization and continuous feature scaling in the EncoderDecoderTimeSeriesDataModule._preprocess_data() method, addressing the TODO comment and completing a critical roadmap item for the PyTorch Forecasting v2 API rework.

Changes

Core Implementation (pytorch_forecasting/data/data_module.py)

1. Target Normalization:

   • Applied target_normalizer (if provided) to normalize target values
   • Supports TorchNormalizer instances with automatic fitting
   • Stores normalization parameters (target_scale) in the result for inverse transformation during prediction
   • Handles both target_normalizer and _target_normalizer attributes (for "auto" case)

2. Continuous Feature Scaling:

   • Applied scalers from scalers dictionary to continuous features
   • Supports both scikit-learn scalers (StandardScaler, RobustScaler) and PyTorch normalizers (TorchNormalizer, EncoderNormalizer)
   • Feature lookup by name (from metadata) or by index
   • Automatic fitting if scalers are not already fitted
   • Proper handling of numpy arrays (sklearn) vs tensors (torch normalizers)

3. Return Type Update:

   • Changed return type from list[dict[str, Any]] to dict[str, Any] (single dictionary)
   • Updated method signature and docstring

Solves #1995