By: ValentinHirschi <valentin.hirschi@gmail.com>
retain LU ray data and account for complete sampling work Keep the represented energy ray with each physical LU root and use it for the original raised eta derivatives. Preserve existing non-LU scalar evaluation and seed conventions; test the deliberate finite-precision routing difference at the new LU boundary. This prepares the native host handoff but does not yet enable cross-section joint channels or adopt sampling roots in physics. Accumulate sampling/source work and actual target-body work across all native attempts and probes. Include summed maps, direct inverse and diagnostic remaps, charge canonical preparation once, and retain failed-attempt costs through rescue. Correct the overlapping event row in benchmark timing. Validation: 174 focused core tests plus the full generated-kite acceptance pass, as do three API acceptance/bookkeeping tests. Core/API check, format, diff and changed-line clippy gates pass. Three native GL638 replays cover two stored points, all 936 orientations and six physical events; totals and events agree with the archived baseline and all 35 state files are unchanged. The portable replay records the exact source snapshot and transparent tuple alias added after linking. No serialized state layout is changed. User prompt: (continue as planned but make sure the time spent in the sampling maps is still quite significantly less than the actual GL638 all-orientation integrand, i.e. it's not crucial to optimize the runtime of the sampling, except if it is somehow comparable to the integrand time; it should be like 10% at the very most, once below that then no need to optimize further).