Increase precision of quantile estimation

This release fixes #129 which reported that quantile estimation in reports looked wrong for some data sets. The issue was the streaming quantile algorithm used wasn't tuned appropriately. However, after considering the problem for a while, it became clear that using such an algorithm is not necessary since we have all the data ready anyways. Therefore, I implemented R's 8th type of batch quantile estimation which is approximately median-unbiased regardless of the data distribution.