feat(optimization): wire up evaluator, problem, and study (M3)

[vcaries]

M3 — Optimization layer (pymoo)

What

• AirfoilEvaluator.evaluate: genome -> params -> geometric prune ->
  XFOIL -> physical constraints -> objectives. Failures (bad genome,
  generator/solver errors) translated to finite sentinel values.
• AirfoilProblem._evaluate: pymoo Problem adapter, population-wide
  evaluation, constraint-aware.
• HistoryCallback: properly inherits pymoo Callback, exposes
  best_per_generation() and __len__.
• OptimizationStudy.run / save_checkpoint / load_checkpoint /
  resume. End-to-end demonstrated on a synthetic L/D objective —
  GA converges to the analytic optimum.

Tests

+12 unit tests (now 117 total). All green:

• ruff / ruff format / mypy / pytest ✅

Out of scope (next)

• M4: campaigns layer (parallel runners, persistent ResultStore)
• M5: visualization (matplotlib plots + GIF/MP4 animation pipeline)