param management inconsistent

[LeonieFreisinger]

Status Quo
A benchmark instance or experiment instance receives various parameters as input. Currently, the input is given by defining parameters in a dataset_list and in a model_classes_and_params list.
Example:

    dataset_list = [
        Dataset(df=peyton_manning_df, name="peyton_manning", freq="D"),
    ]
    model_classes_and_params = [
        (SeasonalNaiveModel, {"n_forecasts": 4, 'K':5}),
    ]

Subsequently, they are assigned to the attributes of different classes and/ or separated into further categories. Currently, this happens partly in the __post_init__() of the Experiment(ABC) class and the __post_init__() of the Model(ABC). So far, we separate the params into _data_params and model_params.

Problem

1. It would be great, to have one dedicated part to post-process, split and assign the input params. In my eyes, this should be on the experiment level.
2. It's great to distinguish between _data_params and model_params. Further, I think we should introduce pred_params, which saves all prediction task-related information.