Add moving average option for ML classifiers (fixes #640)

[samerzumot]

• Implement MovingAverageClassifier in C++ core supporting custom scores and built-in metric models
• Add MOVING_AVERAGE_CLASSIFIER enum across Python, Java, C#, TypeScript, Rust, Swift, Julia, and MATLAB bindings
• Add automated test moving_average_classifier.py

[Andrey1994]

Hi, I was thinking about adding moving average option to already existing classifiers instead of adding a new classifier type.
To enable it or not other_info field can be used.
In terms of implementation will need to think how to make it generic and uniform for all classifier types

[samerzumot]

Hi @Andrey1994,

Thanks for the feedback! I've updated the implementation according to your suggestion:

1. Removed the Standalone Classifier
2. Generic Implementation in BaseClassifier:
   • This makes moving average support completely generic and uniform across all classifier types—supporting both scalar outputs (Mindfulness, Restfulness) and multi-channel vector outputs (ONNX, DynLib).
   • Third-party .dll/.so plugin symbol resolution ("prepare", "predict", "release") in DynLibClassifier remains 100% unchanged.
3. other_info Parsing:
   • Accepts JSON ({"window_len": 4}, {"moving_average": true}, {"period": 5}), key-value (moving_average=5, window_len=4), or simple integer string ("3").
   • Defaults to DEFAULT_MOVING_AVERAGE_WINDOW = 5 when a flag without a window length is provided.
   • Moving average is only activated if recognized keys are found; arbitrary strings or unrelated JSON configs for custom plugins leave moving average disabled to prevent unintended collisions.
4. Memory Bounded: The sliding window history is explicitly trimmed upon every prediction cycle.
5. Tests: Updated moving_average_classifier.py with comprehensive tests covering baseline raw predictions, window smoothing, restfulness score inversion, default fallbacks, key-value syntax, negative tests, and multi-element vector output via DynLibClassifier.

Please take a look when you get a chance!