review and refactor AnomalyLikelihood code (both c++, python)

[breznak]

C++

• needs proper tests ported
• needs bindings for python
• verify the code is actually correct

Python

• improve the code by using library function (ie to compute mean, etc)
• try to minimize the API

Related #124 , review of the code in #457

[breznak]

See numenta/nupic-legacy#3870 for possible update to Likelihood

[breznak]

As discussed in #666 (that number indeed does bring problems! 👿 ) there may be problems in how Likelihood is implemented.

From hinsight, I see 2 generic problems:

1. Likelihood may not handle similar,but not same, values well (needs verification):

anomaly scores: 1.0, 1.0, 1.0, 1.0 -> likelihood (anomalyProbability()) goes down, that is the purpose of AnLikelihood. Same for a sequence of low scores, etc.

I wonder if similarity is handeled by this model? 0.9, 0.99, 0.9, 0.99, 0.9, 0.991337 - 0.991337 is unique (yet same as 0.99), will its likelihood be correctly low?

2. I'd like to compare the current likelihood with its "biological" implementation - implemented as a HTM system. {anScore} -> encode -> SP -> TM -> anomaly (~likelihood)

3. and see if raw should be included? {raw, anScore}