Introduce shadow model training for inference attacks

[beat-buesser]

Description

This pull request introduces shadow model training for inference attacks. This PR is a continuation of #1345 to adress review comments.

Type of change

Please check all relevant options.

• Improvement (non-breaking)
• Bug fix (non-breaking)
• New feature (non-breaking)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

[codecov-commenter]

Codecov Report

Merging #1395 (48a59b0) into dev_1.9.0 (98627f8) will increase coverage by 0.13%.
The diff coverage is 90.54%.

@@              Coverage Diff              @@
##           dev_1.9.0    #1395      +/-   ##
=============================================
+ Coverage      90.43%   90.57%   +0.13%     
=============================================
  Files            232      233       +1     
  Lines          18885    19002     +117     
  Branches        3089     3111      +22     
=============================================
+ Hits           17079    17211     +132     
+ Misses          1042     1030      -12     
+ Partials         764      761       -3     

Impacted Files	Coverage Δ
art/estimators/classification/catboost.py	91.30% <50.00%> (-4.16%)	⬇️
art/estimators/classification/classifier.py	96.49% <60.00%> (-3.51%)	⬇️
...ks/inference/membership_inference/shadow_models.py	90.56% <90.56%> (ø)
...attacks/inference/membership_inference/__init__.py	100.00% <100.00%> (ø)
...ttacks/inference/membership_inference/black_box.py	96.57% <100.00%> (+0.08%)	⬆️
art/defences/detector/evasion/detector.py	74.02% <100.00%> (+1.18%)	⬆️
...fences/detector/evasion/subsetscanning/detector.py	83.80% <100.00%> (+0.63%)	⬆️
...mators/certification/randomized_smoothing/numpy.py	87.09% <100.00%> (ø)
art/estimators/classification/GPy.py	89.09% <100.00%> (ø)
art/estimators/classification/blackbox.py	98.00% <100.00%> (ø)
... and 16 more