Fix std in the summary statistics

[YC-1412]

Hi Guillermo,

This PR fixes #324.

When sample_weight is provided, the ssum calculation needs to multiply the weight after $y^2$ instead of ymask (multiply weight first and then take the square). It is related with #323 at some level (both are about sample weight) but needs a separate fix.

$$\begin{align} std &= \frac{\sum w_i (y_i-\bar{y})^2}{\sum w_i} \\\ &= \frac{\sum w_i y_i^2 + \bar{y}^2\sum w_i - 2\bar{y}\sum w_i y_i}{\sum w_i}\\\ &= \frac{\sum w_i y_i^2 + \bar{y}^2\sum w_i - 2\bar{y}^2\sum w_i}{\sum w_i}\\\ &= \frac{\sum w_i y_i^2 - \bar{y}^2\sum w_i}{\sum w_i}\\\ &= \frac{\sum w_i y_i^2}{\sum w_i} - \bar{y}^2\\\ &= \texttt{np.sqrt(s\_ssums / s\_n\_records - mean ** 2)} \end{align}$$

where s_ssums = $\sum w_i y_i^2$.

I used the test code in #324 to test the fix. Row 3 (group [8.5, 9.5)) now has std=0 as expected.

Before

After

[guillermo-navas-palencia]

Very nice work! Thanks for looking at it.

[guillermo-navas-palencia]

I think this branch must be updated with develop to pass the tests.