EQUIL - BUGFIX - Detect field-line ODE failure instead of consuming a truncated surface

[logan-nc]

The bug

direct_fieldline_int solves the flux-surface field-line ODE and never checks sol.retcode:

https://github.com/OpenFUSIONToolkit/GPEC/blob/develop/src/Equilibrium/DirectEquilibrium.jl#L293-L296

The very next line consumes sol.u[end] as though the field line had closed at eta = 2*pi. But a
failed solve does not throw — it returns a solution truncated wherever the integrator gave up. So
the caller silently builds a flux surface from a partial trace, and the damage surfaces downstream as
unrelated nonsense, or not at all.

The fix

Two checks, because neither subsumes the other:

• sol.retcode != ReturnCode.Success — the ordinary failure.
• sol.t[end] did not reach 2*pi — a solve can stop early through a callback-driven terminate and
  still report Success. direct_fieldline_int installs a DiscreteCallback, so this is reachable.

Both errors name the psifac that failed, and the first points at psihigh sitting too close to the
separatrix, which is the usual cause. The failure becomes attributable to a surface rather than a
stack trace somewhere further down the pipeline.

Verification

• test/runtests_equil.jl — 281/281 pass.
• DIII-D-like ideal example runs end to end (equilibrium → FFS → PE → KineticForces, 476 s), neither
  guard firing.
• Regression harness, diiid_n1, --refs develop,local against develop @ 4e430f170:
  47 quantities, all unchanged, 0.0e+00 across the board. Expected — on a healthy equilibrium this
  adds two predicates and changes nothing.

A note on the test

The added testset asserts the guards against the source text rather than triggering them. A genuinely
non-closing field line is not cheap to synthesize here, and forcing psifac toward the separatrix
would likely hit a different error first and be flaky across machines. The standing positive control
that neither guard false-fires is the EFIT testsets already in that file, which trace real surfaces
through this function. Happy to swap in a behavioral trigger if a reviewer sees a clean way to build one.

Scope

Equilibrium only; no SLAYER/Tearing code touched. Split out of the FKR-width work as an
independent fix, same shape as #399.

---

⚠️ NO MERGE WITHOUT THIRD-PARTY HUMAN REVIEW — NON-NEGOTIABLE ⚠️

🤖 Generated with Claude Code

https://claude.ai/code/session_01PSrf6JCViFfVzqzkQ66o6b