Fixes on feature selection (incremental & smd)

[K1rubel]

This PR introduces a significant refactor and optimization of the Incremental Feature Selection (IFS) module, addressing critical numerical bugs, and performance bottlenecks found in the previous implementation.

Description

Motivation and Context

Optimization and bug fix.

How Has This Been Tested?

Tested locally and awaits more tests i

Enhanced Testing & Integration

Created tests/incremental-feature-selection-test.metta with comprehensive validation across Parity, Majority, and Mux data sets.
Successfully integrated the new logic into the representation module and verified it with updated integration tests in representation-test.metta.
Utility Updates: Patched smd.metta and other helpers to resolve function name conflicts (e.g., updating first to first-from-pair).

Cache Management

LRU Policy: The cache includes a monotonic clock and automatic eviction of the oldest entries when reaching capacity (default 128).
One quirk here is for test case with different tables, cache needs to be cleared after use with a data table so that cache for that table cannot be used for the next one.

Next Steps

Use the entropy and MI calculation for other selection algorithms as well.
Find a general caching implementation to be used for other expensive procedures as well without needing to store big data table as part of the key.

Checklist:

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