By: ValentinHirschi <valentin.hirschi@gmail.com>
Parallelize finite master refinement with exact ordered block merging Share a licensed work-stealing executor for exact source preparation and compact reachable-pivot elimination, preserving original column admission, auxiliary equations, guards and checkpoint cursors. Add observed batch progress, operational window/block settings and refiner-only CPU/worker overrides on both steering lanes. Validate native deterministic source-prefix controls, interruption/resume and 128-worker admission; document repeated five-loop solver speedups and remaining serial phases.