cvx-quadprog: Goldfarb/Idnani dual quadratic programming in NumPy and SciPy

A pure NumPy/SciPy implementation of the Goldfarb–Idnani dual active-set method for strictly convex quadratic programs, requiring no compiler and no build step. It is a reimplementation of the quadprog package — which wraps C descended from Berwin Turlach's Fortran translation — and is API-compatible with it.

The implementation makes design choices the C reference does not contain: constraint insertion uses a single Householder reflection in place of a chain of Givens rotations; columns holding a single nonzero are detected so that bound constraints become indexing rather than reductions; the triangular factor is held in packed storage so the active submatrix stays admissible to a BLAS packed solve; and LAPACK/BLAS kernels replace hand-rolled scalar loops throughout.

The solver was written from the algorithm as published, not from the C. The relationship between the two implementations is recorded in the repository's PROVENANCE.md.