variable_groups doesn't combine real values for scalers

• PyTorch-Forecasting version: 0.9.1

Expected behavior

According to the documentation:

variable_groups (Dict[str, List[str]]) – dictionary mapping a name to a list of columns in the data. The name should be present in a categorical or real class argument, to be able to encode or scale the columns by group. This will effectively combine categorical variables is particularly useful if a categorical variable can have multiple values at the same time. An example are holidays which can be overlapping.

I expect to get equal scalers for multiple columns from the same variable_groups. Or maybe a single scaler?

Actual behavior

Actually, they have different parameters.

Code to reproduce the problem

from pytorch_forecasting import TimeSeriesDataSet
import pandas as pd
import numpy as np

data = pd.DataFrame(
    {
        "group": np.zeros(5),
        "time_idx": np.arange(5),
        "a": np.random.rand(5),
        "b": np.random.rand(5),
        "target": np.random.rand(5)
    }
)

ds = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    group_ids=["group"],
    target="target",
    min_encoder_length=4,
    max_encoder_length=4,
    min_prediction_length=1,
    max_prediction_length=1,
    time_varying_known_reals=["a", "b"],
    variable_groups={
        "var_group": ["a", "b"]
    }
)

assert ds.scalers["a"].mean_.item() != ds.scalers["b"].mean_.item()

Start at TimeSeriesDataSet and trace how variable_groups is handled for time_varying_known_reals and how ds.scalers is populated. Run the provided reproduction to confirm the differing parameters, then clarify whether grouped real columns should share scaling parameters or use one scaler; done means the behavior matches the documented grouping semantics and is covered by a regression test.