Evaluators normalization: more precise range

[lrozenblyum]

At the moment, our evaluators are normalized at some absolute range.
For example see ProtectionEvaluator.

Often there is some obvious (or not hard-to-calculate) relative range to be calculated.

So that we would check the desired characteristic exactly for current position and not from theoretical point of view.

Need to check if that brings more independence between evaluators (which is desired) or worsen it.