1. Rely on actually computed standard deviations in bayestopt_ instead of potentially unset estim_params_
2. Remove arbitrary normalizations/omitted normalization in case of division by 0 in normalization
3. Distinguish between 0 identification and division by 0 due to normalization in plots