mg7 launch: translate madevent run_card names in "set" (cuts, beams, PDFs, scale codes)

[oliviermattelaer]

Targets main (it was stacked on #223, merged 2026-10-10). Three commits plus a merge of main.

Summary

At the mg7 launch question, madevent run_card.dat names that the run_card.dat -> run_card.toml conversion (RunCardMG7.from_LO) already handles were lost. They printed WARNING: invalid set command ... and the run went on with the mg7 default; seen on main 44b8e19, where a p p > e+ e- run integrated with no m_ll cut. Two of them aborted a scripted launch outright. This PR translates them the way from_LO does.

Commit 1: cut names (set mmll 200, set ptj 30, set etaj -1, set ihtmin 100, ...)

• New RunCardMG7.set_lo_cut writes the [cuts] entry from_LO uses. It writes only to [cuts], never [histograms], which shares the key syntax (e.g. sqrt_s.min).
• A minimum <= 0 or a negative maximum removes the bound, as in from_LO (shared _lo_cut_is_active).
• A cut on a _LO_CUT_GROUPS group (parton, alllepton) also adds that group, or the launch fails with KeyError: 'parton'.

Commit 2: beam, PDF and scale settings

• set dynamical_scale_choice 3 / set sde_strategy 2: these used to abort a launch script with InvalidCardEdition ("Please report this bug"). The integer codes and the madevent et/shat/ckkw names now become the mg7 names. Codes with no mg7 equivalent (0 = user-defined, -1 = CKKW) warn and keep the current value. HT and HT/n stay with the generic editor, as before.
• Values translated as in the conversion:
  • ebeam1, ebeam2 (and the set ebeam shortcut, with energy units) → beam.e_cm
  • lpp1, lpp2 (and set lpp) → beam.leptonic
  • pdlabel, pdlabel1, pdlabel2, lhaid → beam.pdf
  • fixed_fac_scale(1/2) → beam.fixed_fact_scale
  • maxjetflavor → the jet group
  • the per-pdg cuts {pdg: value} → a pdg<N> group plus its cut. mxx_only_part_antipart picks pair_mass or sfos_pair_mass, whichever of the two is set first.
• One value for both beams: mg7 has one energy, one PDF and one lepton flag for both beams, so the per-beam settings of one question are combined. ebeam1 and ebeam2 give e_cm = 2 sqrt(E1 E2); lpp1 and lpp2 give beam.leptonic. What still differs per beam is reported once, when the question closes (MG7Selector.check_card_consistency), so set ebeam1 X followed by set ebeam2 X stays quiet.
• Docs: the "Coming from MadEvent" docs page lists what is translated.

Commit 3: per-pdg cuts in the run_card.dat conversion

• A run_card.dat is read with string pdg keys (pt_min_pdg = {6: 100} gives {'6': 100.0}), but from_LO (from Document the mg7 run card (run_card.toml) #223) only took integer keys. So the per-pdg cuts of a real card were dropped without a report. Document the mg7 run card (run_card.toml) #223's test set the dicts with integer keys directly, which hid it.
• _lo_pdg_dict turns the keys into integers (keeping default) before from_LO reads them. New test test_from_LO_pdg_cuts_read_from_a_run_card writes and rereads a real run_card.dat; without the fix it fails with KeyError: 'pdg6-pt'.

Interplay with #154 (one energy and one PDF per beam)

#154 routes ebeam*, lpp*, pdlabel* and lhaid to its per-beam keys through RunCardMG7.set_lo_beam_param (lo_beam_params). Whichever PR merges second should make set_lo_param hand those names to set_lo_beam_param and drop the "one value for both beams" warnings for them. fixed_fac_scale, maxjetflavor, the per-pdg cuts and the integer codes stay here.

Testing

• After merging main (df50882): test_banner.py 85 tests OK, test_mg7_launch.py 93 OK (none skipped, madspace rebuilt), tests/unit_tests/madspin 665 OK. A fresh p p > e+ e- launch with set mmll 200, dynamical_scale_choice 3, sde_strategy 2, ebeam 6800, pt_min_pdg {11: 30} applies them all and gives 1.528 pb.
• Before the merge: tests/unit_tests/various/test_banner.py: 83 tests OK. New: test_set_lo_cut, test_set_lo_cut_composite, test_set_lo_param_value_translation, test_set_lo_param_beams_combine, test_set_lo_param_pdg_cuts, and test_set_lo_param_matches_from_LO (set lands where from_LO puts the same values).
• tests/unit_tests/interface/test_mg7_launch.py: 92 tests OK, none skipped (madspace built from this branch). The new tests stub the generic editor and assert it is never reached for these names, and that mg7 values (transverse_mass, denominators, HT/4) still are.
• tests/unit_tests/madspin: 665 tests OK.
• All 224 run_card.dat parameters were run through the real launch question one by one, before and after this change.
• End to end with bin/madgraph:

launch script	result
p p > e+ e-, set mmll 200, set mmllmax 1000	1.455(14) pb
same with mmll 0 / mmllmax -1 (cut removed)	891.9 pb
p p > e+ e- j, mmll 60	226.3 pb
same plus ihtmin 100, ptllmin 50	13.88 pb
script with dynamical_scale_choice 3 and 0, sde_strategy 2, ebeam 6800, maxjetflavor 4, lpp1/2 1, fixed_fac_scale F, pt_min_pdg {11: 30}, ebeam1 7000, ebeam2 4000	no abort, no "invalid set command"; card has e_cm = 10583, denominators, the 4-flavour jet group and pdg11-pt.min = 30; one asymmetric-beam warning after the question; 503.0 pb

• No Sphinx locally: the edited .rst page was checked with docutils (no warnings).