Add native support for matheuristics (ILS, VNS, GRASP, ...)

[eugenioLR]

Matheuristics combine exact or heuristic subroutines with metaheuristic frameworks. The most common pattern—wrapping an entire optimization algorithm as a local‑search operator—is currently not directly supported: a strategy or algorithm cannot easily be called as a subroutine inside another strategy without loosing all tracking capabilities.

Full matheuristic support would allow users to compose iterative local
search, variable neighbourhood search, GRASP, scatter search, and other
classic hybrid methods out‑of‑the‑box. It depends on constructive
algorithms (#82) for warm‑starting components and will benefit from the
generic component‑swapping infrastructure planned in the EDA refactor (#84).

A partial implementation of VNS was present on previous versions, but it had many problems which made it unworkable for serious applications.

Matheuristics

• Allow an Algorithm (or SearchStrategy) to be wrapped as an
  Operator so it can be used as a subroutine (e.g. local search
  inside a GA)
• Implement greedy initializers that use constructive heuristics
  (reuse infrastructure from Add native constructive algorithms #82)
• Ensure embedded sub‑algorithms can report progress (nested reporters)
  without interfering with the outer loop
• Provide Iterated Local Search (ILS) as a built‑in strategy
• Provide Variable Neighbourhood Search (VNS) as a built‑in strategy
• Provide GRASP (Greedy Randomized Adaptive Search Procedure) as a
  built‑in strategy with configurable constructive phase
• Provide Scatter Search as a built‑in strategy
• Document the pattern for users to build their own matheuristics