[M5][OPTIMIZATION] Implement robust multi-objective therapy optimization and Pareto analysis

[andreazedda]

Objective

Create an uncertainty-aware optimization layer that compares therapeutic research strategies across multiple explicit objectives without producing an unsupported single best treatment claim.

Objective domains

disease-control proxy
residual-disease or relapse proxy
normal-lineage and immune recovery
hepatic, marrow, renal, neurological, and immune toxicity
functional outcomes
interruption burden
feasibility
uncertainty and model risk

Scope

• Pareto dominance and non-dominated set computation;
• robust dominance under parameter, observation, and structural uncertainty;
• sensitivity to objective definitions and weights;
• constrained optimization and infeasibility reporting;
• hypervolume and ranking-stability diagnostics where appropriate;
• utility elicitation kept separate from biological simulation;
• comparison of scalarization, evolutionary multi-objective methods, and transparent baselines;
• reproducible optimization manifests.

Acceptance criteria

• The platform reports Pareto sets before any scalarized ranking.
• Dominance stability under uncertainty is quantified.
• Objective weights and utility assumptions are visible and versioned.
• Infeasible strategies are not silently assigned poor scores and retained.
• Different model structures can change the Pareto front and this is reported.
• Sensitivity and ablation results identify fragile rankings.
• Optimization is reproducible from a fixed candidate set, model, seed, and configuration.
• Runtime language does not claim a clinically superior treatment.

Non-goals

Optimization results are research priorities under explicit models and constraints, not treatment recommendations.