[FIX] chore: explicit in the code the differences between the SDK's `X_df` and the API's `X_df`

[AzulGarza]

The SDK and the API handle the exogenous variables differently. The SDK differentiates between historical exogenous variables (used during finetuning, for example) and future ones. The historical variables are included along the target variable in df, and the future ones are passed in X_df. However, the API receives the exogenous variables stacked together (X_df_api = df_no_target + X_df).

The problem is that the core code needs to apply a proper naming convention to reflect the differences (X_df is used for both cases). This is important because development has been problematic lately (see, for example, #171).