By: ValentinHirschi <valentin.hirschi@gmail.com>
Correct legacy global phases in GL134, GL297 and GL638 inputs Use the default unit graph numerator with the inherited LU phase convention. Migrate the reusable DOT copies and generator, train GL134 on Re, and flag state regeneration in the source cards. Preserve live runs and saved results. Validation: fresh full-orientation GL134 generations at two momentum points rotate the old imaginary answer into Re; the corrected card also generates and passes inspect. Archive commands, outputs and input hashes. Changed TOML parses and DOT phase-only comparisons pass; 34 live-state files are unchanged. User request: Ok, fix these and commit+push (for all three graphs and the corresponding dots), but leave current live runs untouched. We'll refresh them some other times. Confirm that now without custom overall normalisation (i.e. leaving it as `1`) then normalization for GL134 comes out right (i.e. real). Then look into the current weighing factors for GL638 you came up with (in the .dot file) and explain what they achieved (i.e. what is their concise expression in shortened mathematical notation) and what kind of additional primitives would be needed on top of eta surfaces to be able to write these functions more concisely? Follow-up: (did you commit+push the small revision with `1` for the overall graph numerator everywhere)