raise ValueError in remove_nan(), if data is NaN only
When remove_nan() is evaluated no error message is raised, when there is a timeseries (history, label, observation) that only consists of NaN. Until now all timeseries are overwritten by None. This leads to errors that are raised afterwards (for instance self.history.transpose in get_transposed_history()).
Suggested fix is to raise error after the non_nan_xxx declaration with a definitive message.