Migrate off GSL onto the fortnum numerical core

[krystophny]

Supersedes #135.

Addresses #134. KAMEL is MIT-licensed but linked GSL (GPL-2.0+). This PR removes GSL entirely and routes every numerical call through the fortnum library (fetched from lazy-fortran/fortnum), the single numerical core for the group. Where #135 reimplemented the GSL routines in-tree and pulled Bessel from libneo, this PR depends on fortnum directly. No shipped binary links GSL.

What moved

KIM (Fortran)

• gsl_sf_bessel_In -> fortnum bessel_in in FLR2_asymptotics.f90 (negative orders via I_{-n}=I_n, handled inside bessel_in).
• gsl_sf_erf -> the Fortran 2008 erf intrinsic (integrands.f90, integrands_adaptive.f90, numerics_utils.f90).
• gsl_mod.f90 deleted (the gsl_sf_erfc/gsl_sf_dawson bindings were dead). KIM_lib links the Fortran fortnum target; find_package(GSL) and the GSL link/include are gone.

KiLCA / QL-Balance (C/C++), via the fortnum C ABI (fortnum.h)

• gsl_odeiv rk8pd evolve loops -> fortnum_ode_integrate_dop (calc_back.cpp, incompressible.cpp, compressible_flow.cpp). The historical rhs_* callbacks forward through small fortnum_ode_rhs adapters; the returned mesh is replayed exactly as the evolve loop stored each accepted step.
• gsl_integration_qag -> fortnum_integrate_qag (four_transf.cpp, calc_I_array_drift.cpp); gsl_integration_qagi -> fortnum_integrate_qagiu with inf=2 (vel_integral.cpp).
• gsl_root_fsolver brent -> fortnum_root_brent (calc_mode.cpp).
• gsl_multiroot_fdfsolver hybridsj -> fortnum_multiroot_hybrid (calc_eigmode.cpp); eval_det is reexpressed on the fortnum_vector_fn ABI and the finite-difference Jacobian moves into the solver. The dead qng-based calc_circle_integral_gkq and the gsl_multiroot print helpers are removed.
• gsl_deriv_central -> fortnum_deriv_central (incompressible.cpp, compressible_flow.cpp).
• gsl_heapsort_index -> fortnum_argsort via a sort_index_doubles helper in shared.cpp (calc_cond.cpp, sysmat_profs.cpp, adaptive_grid.cpp).
• KiLCA hyper1F1 quadrature path -> fortnum_hyperg_1f1_a1; the GSL-only Levin variant (marked "do not use" upstream) is removed.
• Public interfaces (rhs_*, jac_*, the Afunc/Bfunc helpers) are preserved so callers compile unchanged; only the GSL internals are replaced.

Build

• cmake/FetchFortnum.cmake fetches fortnum via FetchContent from git@github.com:lazy-fortran/fortnum.git (GIT_TAG main), guarded by if(NOT TARGET fortnum) so a transitive pull (e.g. through libneo) does not double-declare the target. The Fortran fortnum target backs the Fortran sources; ${fortnum_SOURCE_DIR}/include is on the C/C++ include path for fortnum.h.
• cmake/FetchGSL.cmake deleted and include(FetchGSL) removed; gsl/gslcblas dropped from the KiLCA, KIM, and QL-Balance link lists and include paths.
• KIM_exe and the KIM KiLCA-dependent tests link the freshly built kilca_lib target instead of a hardcoded stale prebuilt archive under build/.
• GSL removed from the setup scripts, the standalone KiLCA build helpers, and the dependency lists in the READMEs.

This depends on the fortnum C ABI forwarding a user void* ctx to the callbacks (fortnum_integrate_*, fortnum_root_brent, fortnum_deriv_central, fortnum_multiroot_hybrid, fortnum_ode_*); that small ABI addition is committed on fortnum main (dc277b5).

Verification

Linux x86_64, gfortran/gcc 16.1.1, CMake 4.3.3, Ninja, fortnum main. All dependencies (LAPACK, SuiteSparse, SUNDIALS, NetCDF, HDF5, fortnum) resolved; no GSL fetched or built.

Before (origin/main): cmake/FetchGSL.cmake fetches and builds GSL 2.4; KiLCA/KIM/QL-Balance link gsl/gslcblas.

After (this branch): full build succeeds, no GSL symbol in any shipped binary.

$ cmake --build build -j  (forced recompile of all migrated sources)
[16/28] Linking CXX static library install/lib/libKiLCA_Lib_..._Release_64bit.a
[22/28] Linking CXX executable install/bin/KiLCA_Normal_..._Release_64bit
[23/28] Linking CXX executable install/bin/KIM.x
[28/28] Linking CXX executable install/bin/ql-balance.x
EXIT=0

$ for b in KIM.x KiLCA_Normal_... KiLCA_EigParam_... ql-balance.x post_processor_64bit fouriermodes.x; do nm install/bin/$b | grep -ic gsl_; done
KIM.x                       gsl=0
KiLCA_Normal_...            gsl=0
KiLCA_EigParam_...          gsl=0
ql-balance.x                gsl=0
post_processor_64bit        gsl=0
fouriermodes.x              gsl=0

ctest (18/18 pass), including the KiLCA-math tests that now run through fortnum:

$ ctest
 1/18 test_rk4_integrator ............. Passed
 ...
10/18 test_integration_methods ....... Passed
11/18 test_quadpack_qagi ............. Passed
12/18 test_root_finding .............. Passed
15/18 test_ampere_matrices ........... Passed
17/18 test_rhs_balance ............... Passed
18/18 test_kim_adapter ............... Passed

100% tests passed, 0 tests failed out of 18

fortnum's own suite (67/67) passes against the C-ABI ctx change, including a new context-forwarding check in the C smoke test:

integrate_qag(ctx scale*x)   got=1.5 want=1.5 abs_err=0 PASS
ALL CAPI SMOKE CHECKS PASSED
100% tests passed, 0 tests failed out of 67

Remaining

• The bundled AMOS complex Bessel (zbesj/zbesi/zbesk in hmedium.f90, calc_D_array_*.f90, bessel.f90, plus the KiLCA/math/bessel/amos tree) is not GSL and is untouched here. Migrating it to fortnum bessel_j_complex/bessel_i_complex/bessel_k_complex and deleting the bundled AMOS is a separate follow-up; it does not affect the GSL removal.
• KIM's own dispersion sources (fun_input.f90, wkb_dispersion.f90) call zbesi from the bundled SLATEC (not GSL); also untouched.

[krystophny]

Superseded by a stack of eight atomic PRs that re-cut this migration off origin/main and also complete the AMOS/SLATEC removal this PR left as follow-up. Each is individually reviewable; merged in order they compose to the full migration. A human should close this PR.

• Migrate KIM special functions to fortnum (1/8) #140 (1/8) KIM special functions: gsl_sf_bessel_In -> fortnum bessel_in, gsl_sf_erf -> intrinsic erf; FetchFortnum at a7faa3c. base: main
• Migrate KIM Fokker-Planck QUADPACK to fortnum (2/8) #141 (2/8) KIM Fokker-Planck QUADPACK -> fortnum integrate_qag/qags. base: Migrate KIM special functions to fortnum (1/8) #140
• Migrate KiLCA/QL-Balance quadrature to fortnum (3/8) #142 (3/8) KiLCA/QL-Balance C/C++ GSL QAG/QAGI -> fortnum_integrate_qag/qagiu. base: Migrate KIM Fokker-Planck QUADPACK to fortnum (2/8) #141
• Evolve KiLCA background and imhd ODEs continuously with fortnum rk8pd (4/8) #143 (4/8) KiLCA ODE rk8pd -> fortnum_ode_integrate_dop. base: Migrate KiLCA/QL-Balance quadrature to fortnum (3/8) #142
• Migrate KiLCA roots, sort, multiroot to fortnum (5/8) #144 (5/8) KiLCA roots/sort/multiroot -> fortnum_root_brent/argsort/multiroot_hybrid/deriv_central. base: Evolve KiLCA background and imhd ODEs continuously with fortnum rk8pd (4/8) #143
• Replace bundled AMOS complex Bessel with fortnum (6/8) #145 (6/8) Bundled AMOS zbesj/zbesi/zbesk and SLATEC zbesi -> fortnum complex Bessel via an AMOS-ABI shim; delete the amos/ and slatec z*.f trees. base: Migrate KiLCA roots, sort, multiroot to fortnum (5/8) #144
• Migrate KiLCA 1F1 to fortnum (7/8) #146 (7/8) KiLCA 1F1 -> fortnum_hyperg_1f1_a1. base: Replace bundled AMOS complex Bessel with fortnum (6/8) #145
• Drop GSL and finalize fortnum wiring (8/8) #147 (8/8) Drop GSL (FetchGSL, gsl/gslcblas), finalize wiring; ddeabm slatec link. base: Migrate KiLCA 1F1 to fortnum (7/8) #146

Verification (stack tip, all eight applied): full CMake/Ninja build green, ctest 18/18, and nm over every shipped binary shows 0 AMOS implementation symbols and no GSL symbols from KAMEL code. The only residual GSL is two gsl_sf_gamma* symbols pulled transitively from libneo, not from KAMEL.

[krystophny]

Split into atomic per-aspect PRs #140-#147 (KIM specials, KIM QUADPACK, KiLCA quadrature, ODE, roots/util, complex Bessel AMOS/SLATEC removal, 1F1, plumbing). Closing in favor of the atomic stack.