Add multi-objective supplier sourcing example (QP frontier + duals)

[cafzal]

Multi-objective supplier sourcing example — companion to the cuopt-multi-objective-exploration skill

A new supplier_selection/ example demonstrating multi-objective Pareto-frontier exploration with cuOpt on a procurement decision — the workflow packaged as the cuopt-multi-objective-exploration skill (NVIDIA/cuopt#1355). It completes the multi-objective family alongside the #151 examples (finance QP, workforce MILP) with a supply-chain QP.

supplier_selection/supplier_sourcing_frontier_duals.ipynb (QP) — one order is split across suppliers when reliability and diversification conflict: the reliable suppliers cluster, so demanding higher reliability concentrates the order. Minimizes concentration risk wᵀ C w (a dense quadratic, built via QuadraticExpression(qmatrix=...)) subject to a fully-allocated order, a per-supplier cap, and a unit-cost budget; traces the concentration-vs-reliability frontier by ε-constraint (sweep the reliability floor) and reads each point's dual — the sensitivity d(concentration)/d(reliability), the diversification given up per added point of reliability.

Reuses the repo's notebook idiom (GPU check → cuopt-cu12 install → solve), ships output-stripped (repo convention — every cuopt-examples notebook has 0 cell outputs; run evidence below), with self-contained synthetic data.

User testing

Built by following the skill, then run end-to-end on Colab T4, cuOpt-cu12 — clean, no API errors:

• The ε-constraint sweep traces an 18-point frontier (reliability floor swept up to the achievable ceiling); all 18 solved to Optimal (0 PrimalFeasible). A single solve was only ever one point on it.
• Active suppliers drop 12 → 6 as the reliability floor rises — multi-sourcing erodes as you demand reliability (shown by the point color, left panel).
• The reliability floor's dual rises 0.103 → 10.348 as the floor climbs (the right panel) — the marginal diversification cost of reliability, steepening into the diminishing-returns / knee region.

Notes

• Synthetic/toy data — 12 suppliers across 3 regions; this demonstrates the method, not a sourcing study.
• Any non-Optimal (e.g. PrimalFeasible) frontier point is flagged in the sweep output; none occurred here.
• Builds the dense concentration quadratic from the matrix (QuadraticExpression(qmatrix=...)) rather than a term-by-term loop — the scalable pattern for a dense Q.

Companion to the cuopt-multi-objective-exploration skill (NVIDIA/cuopt#1355).

[cafzal]

@rgsl888prabhu new multi-objective example – supplier-sourcing QP, concentration-vs-reliability frontier + duals; same shape as the #151 examples. Ready for review.