Feature/woa update feature

[SeyitKaanGunes]

Summary

• Update OriginalWOA to use per-dimension coefficients and a normalized progress schedule.
• Add WOAmM variants that combine a modified mutualism phase with standard WOA moves, plus optional stagnation restart and boundary handling.

Detailed Changes

• OriginalWOA.evolve now computes progress with (epoch - 1) / (self.epoch - 1) (safe for epoch=1), derives a/a2 from that schedule, and samples r1/r2 per dimension. This makes A/C vectors and applies them per dimension inside the update loop, which aligns with the standard WOA formulation.
• OriginalWOA updates each agent with update(solution, target) to keep the solution and fitness consistent.
• Introduces WOAmM, based on “WOA with Modified Mutualism”:
  • Mutualism phase: for each agent, pick two peers; choose the best among the three for X_bf, and update Xi and the worse of {Xm, Xn} using benefit factors (1–2) and random weights (scalar or vector).
  • Optional mutualism_snapshot to read from a snapshot while updating, avoiding in-place interference.
  • Refreshes g_best before the WOA movement.
  • Standard WOA phase uses per-dimension A/C like OriginalWOA.
  • Optional stagnation restart: if the best fitness doesn’t improve for stagnation_epochs, regenerate the worst restart_ratio fraction of the population.
  • Boundary handling via boundary_mode (clip, reflect, random) using amend_solution.
• Adds WOAmM_Paper, a variant documented to follow the paper’s population replacement behavior while keeping the same two‑phase loop.

New/Updated Parameters

• mutualism_vector_rand: use per‑dimension random weights in mutualism updates.
• mutualism_snapshot: snapshot population for mutualism reads.
• stagnation_epochs: epochs without improvement before restart.
• restart_ratio: fraction of worst agents to restart on stagnation.
• boundary_mode: clip (default), reflect, or random.