49d3cc84

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

group threshold counterterms by defining edges and signed cycles

Require complete parent_lmb metadata and preserve native frames while sharing
complete overlap, center and multichannel catalogues across compatible cuts.
Validate explicit group refinements and raised residue consistency. Resolve
grouped-amplitude metadata in the master topology and reject explicit metadata
on non-master cross-section graphs. Fix maximal-overlap catalogue coverage and
preserve native default amplitude geometry and export roundtrips.

Migrate fixtures and GL297 parents. GL638 already supplies full parents; retain
its numerical prescription and document the confirmed coupled ordinary-pole
failure. The broader ordered subtraction remains a pending physics decision.

Validation: 29 final focused tests passed; six fixture tests passed unchanged
in a serial initialized process; workspace all-target check, targeted all-target
clippy with warnings denied, formatting and CLI build passed. GL297: 162 finite
f128 soft samples restore the tested baseline powers. GL638: 216 finite soft
samples are integrable on the tested rays, but an 18-point ordinary-threshold
scan retains a pole. GL638 scans disable integrated UV generation after its
independent failure was reproduced with and without threshold metadata.

Saved-state manifest version is now 7; regenerate older gammaloop_state/ data.

User prompt:
Ok, I assume you've already done some minimal focussed testing (and updated the dot files of the two graphs considered) so you should commit+push to the remote version of that branch.

I now want to understand better the situations with GL638, can you summarize to me (in some details still, with explicit mention of the threshold and weighing functions etc.. your ocnsidered) the situation as well as the problem you identified there (and what escaped the analysis done in ttH\_defo.pdf)?

Timeline