Avalanche Tests: add average of measured biases

[avaneev]

From my experience, the average of all measured biases is an important metric when comparing hash function (at a given bit-size, of course), it's a very useful test that reflects overall "quality" of hash function.

[rurban]

Agreed

[avaneev]

This is useful information, but how to evaluate it is a complex question. E.g. I had two variants of hashing code that showed 0.71% average, but while one gave 1.5 MomentChi2, another gave 6.0 MomentChi2... So, I would add "standard deviation" of these max bias measurements as well - this factor also differentiates hashing algorithm. I think it's not bad to have a high average if std.dev is lower.

[rurban]

No, still planned