d957817a

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

pin the Symbolica merge fix and record the GL638 investigation

Pin Symbolica, Numerica and Graphica to the published minimal constant-merge
backport a277a8dadbaae5a8d6605f2f381991a601e1f345, retaining the current API and
nested tensor-sum counter fix. Keep builtin exact pi in threshold prefactors.

Supply the all-orientation GL638 WH/WF metadata and common run-card blocks with
named generation, sampling, stability and symmetric localization controls.
Record matched full-UV soft-limit controls, signed/absolute integration results,
20-core sampling comparisons and portable numerical counterexamples. Preserve
the explicit limitations at coincident A/P roots and the hard H/Z corner;
ordinary-soft cancellation and improved finite-run errors are not a full cure.

Validation with the public pin: formatting, unchanged Hakari output, locked
workspace/all-target check, 12 focused nextest tests, focused Clippy with warnings
denied, and release build pass. Fresh all-936-orientation generation and a
3,000-sample pilot pass; the pilot has no NaN or unstable samples. Both diagnostic
panels (13 freshly generated state cases and 19 prior-state controls) reproduce
previous numerical payloads and precision flags exactly, including the known
invalid A/P evaluations. Read-only states and compiled inputs remain unchanged.
Card: examples/cli/epem_a_ttxh/NNLO/epem_a_tth_NNLO_test_GL638.toml
Regenerate saved integrands for the DOT/exact-pi changes; runtime controls do not
require regeneration. No saved state or private research archive is committed.

User request:
alright, then publish your patch for now and pin against that gammaloop version. I want you to be able to commit your resulting investigation.
So go ahead and do that but also summarize for me the status of GL638, I'll restart the goal then.

Timeline