Make KiLCA background ODE reproduce the GSL-recorded golden bit-for-bit

[krystophny]

Summary

The QL-Balance golden record (ql-balance_golden_record) compares the
fortnum-backed build against the GSL-era golden at rtol=1e-8/atol=1e-15. The
KiLCA linear-response solve amplifies a ~1-2 ULP background-field difference by
~1e7, so dqle*/dqli*/Br_* diverged from the golden by ~2.7e-6 and the test
failed at every time step.

Root cause: two fortnum kernels did not reproduce their GSL predecessors.

1. The background equilibrium ODE in KiLCA/background/calc_back.cpp is driven
   through the old gsl_odeiv API (gsl_odeiv_evolve_apply +
   gsl_odeiv_control_y_new). fortnum's rk8pd evolve loop modelled the odeiv2
   semantics: it scaled the failure ratio by the pre-step solution and rejected
   every step with rmax > 1.1 unconditionally. The old API updates y in place
   before the controller runs and keeps an otherwise-rejected step when the
   suggested decrease would not move t by at least one ULP. That mismatch
   produced a different accepted-step mesh and a ULP-level background profile.

2. fortnum's deriv_central used a different 5-point stencil than
   gsl_deriv_central and skipped GSL's optimal-stepsize second pass, diverging
   by up to ~1e-10 on smooth functions.

Both are fixed in fortnum; this PR only bumps the fortnum pin. No golden
regeneration, no tolerance change, no skipped cases.

Verification

Before (fortnum at the previous pin):

test_golden_record.py::test_golden_record FAILED
[FAIL] /f_6_2/LinearProfiles/0/Br_Re   max rel 2.722e-06
[FAIL] /f_6_2/LinearProfiles/0/dqle11  max rel 2.705e-06
... 100+ failing quantities across all time steps

After (fortnum at the bumped pin):

============================= 115 passed in 33.57s =============================
worst_abs=0.000e+00 worst_rel=0.000e+00   (cur output bit-identical to golden)

Clean-room check, old gsl_odeiv rk8pd + control_y_new(1e-16,1e-16) vs
fortnum on a stiff scalar ODE carried across a grid: max relative gap 1e-10
before, 0.0 after.

gsl_deriv_central vs fortnum_deriv_central on smooth functions: max
relative gap 2e-10 before, 0.0 after.

Remaining KAMEL ctest cases (special functions, QUADPACK QAGI, integration
methods, KIM adapter, sparse, ampere matrices): 19/19 pass. fortnum's own suite:
74/74 pass.

Note: fortnum pin updated to current main (92de6e9) after a fortnum history rewrite; old shas no longer resolve.

[krystophny]

Superseded by #143. The winning state from this branch (calc_back evolved as one continuous fortnum rk8pd run via the C ABI, plus the imhd incompressible and compressible-flow zones moved from dop853 to the same continuous rk8pd) has been consolidated into the single k4 PR #143. #143 now carries:

• Replace GSL rk8pd ODE evolve in KiLCA with fortnum dop853 (the base step)
• Evolve KiLCA background ODE continuously with fortnum rk8pd (calc_back)
• Evolve KiLCA imhd zones with GSL-faithful rk8pd instead of dop853

The downstream stack (#144..#150) is rebased on top of this consolidated k4, so the continuous-rk8pd background and imhd state flows through the whole stack. The k4 tip reproduces the ql-balance golden record 115/115 (rtol=1e-8, atol=1e-15). fortnum is pinned at 92de6e9 on every branch. Closing in favor of #143.