[Core][Bug] Unbalanced base predictions fail to reshape forreconciler_args

[kdgutier]

Currently, HierarchicalForecast methods require the base predictions to be "balanced", i.e., all the predictions must be available for all the series and forecast creation dates.

If the base forecast inputs are not balanced, the following reshapes fail:

y_hat = Y_hat_df[model_name].values.reshape(len(S_df), -1).astype(np.float32)

y_insample = Y_df['y'].values.reshape(len(S_df), -1).astype(np.float32)

The is_balanced parameter is already available, but a Raise.Exception to check the input characteristics can help.

[elephaint]

I've seen this issue too - this also ties in to #290 - it seems the checks / required data should be method-specific.