f446d6ce

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

retain one sampling draw across physical precision rescue

Prepare cube sampling and selected direct partitions once at a fixed Arb
budget, retaining graph-parent points, combined map/partition factors and
selected cut kinematics. Each physical precision materializes directly from
that anchor. Remove the old unprepared summed variants and repeated native
map traversal, preserving event groups and applying sampling factors once.

Matching discrete policies alone did not guarantee one proposal law: switching
between nearby continuous maps based on the physical result can leave gaps even
when their Jacobians and densities agree. Retain the complete draw instead.
Embed direct binary64 momentum input exactly, check materialization range and
factor accuracy, and isolate row/cache identities across failed probes.

Allow sampling-only energy aliases only after exact signed spatial routing is
certified in the supplied external frame. Keep structural raised-edge equality
and original energy equations unchanged. Hosted joint binding and GL638 timing
remain follow-up gates; no bounded-weight or performance improvement is claimed.

Validation details are recorded in ADVANCED_SAMPLING_PLAN.md and
docs/research/advanced_sampling/CANONICAL_DRAW_SOURCE.md. The saved-state
acceptance retains its numerical criteria with a measured runtime allowance.
No saved-state layout changes.

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