Optimization of Stochastic Maximum Dependency Feature Selection

[K1rubel]

This PR refactors the Stochastic Maximum Dependency (SMD) algorithm and integrates it into the main feature selection dispatcher, improving overall performance in PeTTa environment.

Description

Refactored the SMD feature selection algorithm to use PeTTa's native cons list data structure. Integrated cached-mi-score across the selection process.
Refined the featureSelection entry point in feature-selection-main.metta. The SMD branch now directly calls the optimized implementation, removing legacy manual calculations and ensuring a clean flow for both empty and pre-populated initial feature sets.

Motivation and Context

PeTTa optimizing of the existing code base.

How Has This Been Tested?

Updated tests/smd-test.metta with comprehensive benchmarks for mux6 and majority5 data sets.
Added a specific test case confirming that forced features (e.g., (0 2)) are correctly present in every resulting feature set.

TO DO

For cleaner code, move all the refactored feature selection specific helpers to a separate file, feature-selection-helpers.metta.

Types of changes

• Optimized implementation.
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• My code follows the code style of this project.
• I have added tests to cover my changes.
• All new and existing tests passed.

[Yagth]

I will merge this after you address the import comment.