A2: Pseudo basic steps — Lagrangian dual bounds for GDP #4

Papageorgiou & Trespalacios, EURO J. Comput. Optim. 6:55-83 (2018) simulate basic steps via Lagrangian relaxation of the disaggregated linking constraints: bound improvement without the exponential model growth of true basic steps, and embarrassingly parallel. Unimplemented in any open-source package. (Slide 7: the constraints look like nonanticipativity constraints — the PySP analogy from 2020 now maps to mpi-sppy/PH machinery.)

Scope:

• gdp.pseudo_basic_step-style transformation or a dual-bound utility for GDPopt (targets=[...], lambdas=[...]).
• Multiplier update loop (subgradient/bundle); parallel subproblem evaluation.
• Use as rigorous dual bound provider for LOA/LD-SDA runs that currently report none.

Ref: https://doi.org/10.1007/s13675-017-0088-0 (arXiv:2501.15345).

Part of roadmap epic #2 (item A2).