[RFC][ENH] Outliers: Widget upgrade

[VesnaT]

Issue

Current Outlier widget usually does not find desired amount of outliers.

Description of changes

• wrap sklearn classes for outlier detection (LocalOutlierFactor , IsolationForest)
• introduce separate editors for each outlier detector
• output annotated data (appended Outlier feature)

Things to discuss

• the widget could accept distance matrix and use it for LocalOutlierFactor
• custom method (identify outliers using distance from the mean (threshold could be some multiple of std) for each feature)
• responsive widget: the widget could make calculations in a separate thread to make Orange responsive, but it is impossible to terminate the calculation. The progress could be shown only during preprocessing, before fitting, after fitting, after predicting and during output construction.
• save fitted model into annotated data (Outlier feature's computed value) for novelty detection using apply domain widget

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

Merging #4338 into master will increase coverage by 0.07%.
The diff coverage is 98.85%.

@@            Coverage Diff             @@
##           master    #4338      +/-   ##
==========================================
+ Coverage   86.84%   86.91%   +0.07%     
==========================================
  Files         396      396              
  Lines       71782    71975     +193     
==========================================
+ Hits        62338    62557     +219     
+ Misses       9444     9418      -26