Add robust static/dynamic initialization schemes and population initi… - #14

This commit expands initialization support for static/dynamic hyper-heuristic execution and adds coverage for the new population initialization strategies.

What was done

• Added validated initial-scheme handling in Hyperheuristic, including compatibility with the legacy typo key (intial_scheme) and normalized scheme parsing.
• Propagated initial_scheme through static and dynamic solve paths, and through candidate evaluation.
• Hardened candidate decoding to support NumPy integer types and reject empty encoded sequences with a clear error.
• Updated dynamic operator application to accept direct operator tuples.
• Hardened Metaheuristic initialization when search_operators is None or a NumPy array.
• Extended Population initialization with additional schemes: lhs, sobol, halton, beta, normal, lognormal, exponential, and rayleigh.
• Kept neural-network-specific flow unchanged in this patch.

Where it was done

• customhys/hyperheuristic.py
• customhys/metaheuristic.py
• customhys/population.py
• tests/test_population_initialization.py

How it was done

• Introduced an explicit set of valid schemes and a selector/normalizer method in Hyperheuristic.
• Passed the selected scheme into Metaheuristic runs used by static/dynamic modes.
• Added tuple-aware execution logic in Metaheuristic.apply_search_operator.
• Implemented additional sampling methods in Population, including quasi-random (Sobol/Halton), stratified (LHS), and distribution-based initializers.
• Added focused tests covering output shape/range/finite values, LHS stratification, Sobol size behavior for non-power-of-two agent counts, distributional bias checks, scheme normalization, and unknown-scheme fallback behavior.

Validation

• ruff checks passed for modified files.
• Targeted tests passed:
  • tests/test_population_initialization.py
  • tests/test_metaheuristic.py