MonotonicGA & Lifecycle Hooks

[francastagna]

PR Summary

We implemented changes to AbstractGeneticAlgorithmandMonotonicGeneticAlgorithm`:

• Lifecycle hooks: generation/step observer callbacks to trace GA execution.
• Monotonic GA: enforces pair-wise monotonic replacement and starts each generation from elites.

---

Main Changes

• Base GA (AbstractGeneticAlgorithm):
  • Added lifecycle methods: beginGeneration, endGeneration, beginStep, endStep.
  • Observer API: onGenerationStart, onStepStart, onStepEnd, onGenerationEnd(population, bestScore).
  • score function: now always uses frozenTargets when present; falls back to combined fitness if empty.
• MonotonicGeneticAlgorithm:
  • Starts next population with elites (formTheNextPopulation).
  • Uses the same lifecycle hooks as the base GA.
• GA Recorder/Observer:
  • Introduces observer hooks to record lifecycle and operator metrics.
  • Records bestFitnessPerGeneration using the emitted bestScore (max score of the generation).

Tests

• Monotonic GA tests added:
  • Elites + Monotonic replacement (pairwise strict improvement using score).
  • Edge cases: xoverProbability = 0 (no crossovers) and fixedRateMutation = 0 (no mutations).
  • Monotonicity across generations: assert non‑decreasing bestFitnessPerGeneration (generation score).