Fix for calc_inv_vol_weights() in case of zero volatility

[vfilimonov]

If some returns series supplied to calc_inv_vol_weights() are equal to zero, then inverse-volatility will be infinite.

I suggest to remove assets with zero volatility from the portfolio rather than returning a list of zeros (sum of all weights will be also infinite -> individual weights of good assets are equal to zero).

There're three options:

• we can drop these assets (dropna() here)
• we can set their weights to zero (fillna(0) here)
• we can return NaN for them

I think the last variant is better: user might expect series of the same structure as the initial dataframe with returns, so dropna() is not good. Also it is consistent with the following situation: if we pass a series of NaN then this weight will be equal to NaN

[pmorissette]

Looks good to me - thanks!