[rogeriojorge]

Fix boundary-mode selection, the QS/iota objectives, and flux-matched VMEX export directly on gftd/single_stage at 67f26c4. No merge from main.

The original example keeps its structure and top-level globals, with no argument parser. Changes are limited to three physics modules, the example, focused regression tests, and two small initial-condition files. Loss infrastructure, ALM, dynamics, dependencies, workflows, and other examples are unchanged.

The active objective values and full gradients match the validated implementation exactly at the initial state and saved mode-1/2/3 states. All 64 targeted field, boundary, flux, objective, and derivative tests pass. Fresh-run and restart checks pass. Reloading mode 3 through the globals-only example runs VMEX successfully: edge iota 0.409839, aspect 3.63559, and nine diagnostic figures. The same saved state has independently checked equilibrium QS mean-square 4.53e-5 and coil RMS B·n/B 3.66e-4.

CI: Python 3.11 passes. Python 3.12 fails four unchanged Boozer-compaction tests on roundoff differences of 2.22e-16. Main's separate test-tolerance fix is excluded from this PR.

@gftd13 please review and test the boundary-mode globals, iota-fit safeguard, and VMEX export with your inputs. For continuation, edit RESTART_FROM, restart_filepath, output_filepath, and SURFACE_MPOL/NTOR. At mode 3 use QS_WEIGHT=1e4 and IOTA_MPOL=IOTA_NTOR=8.

[rogeriojorge]

@gftd13 I reduced this to the required fixes directly on your branch, with no main merge or argument parser. The example retains its globals and structure. Objective values and gradients match the validated results; the globals-only reload also reproduces the VMEX edge iota of 0.409839. Please review and test the boundary-mode controls, iota-fit safeguard, and VMEX export with your inputs.