ea6c04a7

By: ValentinHirschi <valentin.hirschi@gmail.com>

share prepared cut kinematics with conditional sampling

Conditional maps and physical LU evaluation now share an authenticated native
ray and root. Each generating row freezes its selected source records before
inverse-density work; only identical plans and represented prerequisites reuse
a preparation. Precision attempts rebuild native records from the original
source, while debug/norm/canonical work keeps separate root histories.

Validate the selected graph, channel, parent and host definition before physical
adoption. Directed checks bound represented root, first-derivative and radial
volume agreement, including rotated probes. Keep raised eta jets on the retained
ray and count adoption validation as sampling overhead. Remove obsolete wrapper
precision casts. No saved-state layout changes.

Validation: 184 unique core tests, including the full generated amplitude
acceptance test, actual conditional/raised physics, affine roundoff, native
replay, successful physical rotation and invalid-host rejection; three API
regressions; cargo fmt/check/clippy and diff checks. Clippy has no new warnings.
GL638 joint binding, stronger pointwise density alignment and the warmed 10%
sampling-cost measurement remain follow-up work in ADVANCED_SAMPLING_PLAN.md.

User request:
(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).

Timeline