Use nLTT::nltt_stat instead of approximation

[richelbilderbeek]

Sure, it was useful to add get_nltt_values to the nLTT package. But now, I could just calculate the exact nLTT statistic instead.