Anomaly detection boilerplate

[bakhtos]

Fixes #15 Only accept this PR after closing #18.

Many train_MODEL functions in AnomalyDetection are refactored away. Instead, a single function train_model accepts a model class and keyword arguments that should be passed to the model constructor. train_model populates the keyword arguments with the default values that were used in train_MODEL functions. This allows to pass any model to the AnomalyDetection class, as well as give any kwargs to models that were already implemented before.

[bakhtos]

Reset this PR to the situation where train_MODELNAMEfunctions are still available and existing scripts using them would not break AND train_model can now accept arbitrary model class and its kwargs while retaining backwards-compatibility with some scripts that could have used the previous version.

This version is only 3 commits ahead of #18

[bakhtos]

The PR is ready.

• train_model function now accepts a model class and its kwargs
• The train_* functions use the new way to call the train_model function.
• The unused parameters in train_*functions are cleaned up`
• evaluate_all_ads uses a set comprehension internally to iterate over all train_* functions.
• Solved #15