Add statistical test flag on G2 in vocab cooc

[Yomguithereal]

https://archimede.mat.ulaval.ca/stt1920/STT-1920-Loi-du-khi-deux.pdf

Investigate weird correlation between G2 and chi2 in full version.

I suspect the self-loops are accouting for most of the weirdness.

g2 should return an Option

let observed_12 = (x - xy) as f64;
let observed_21 = (y - xy) as f64;
let observed_22 = (n - (x + y) + xy) as f64;

12 and 21 can be 0

22 can also be 0 and be a problem with self loops?

• Fix metrics
• drop --complete, make full computation the default
• add --approx for non --distribg
• add test on chi2 for vocab doc-token
• add test on chi2 or g2 on vocab cooc
• add lgl specificity on G2
• avoid self loops altogether in distributional