Add SIMSOPT MHD-energy Optimizable

[krystophny]

Wraps GVEC as a SIMSOPT Optimizable for the fixed-boundary MHD energy.

• GvecEnergyProblem re-solves the interior equilibrium for a boundary vector with a target-preserving scipy hybr root solve and caches the solved state per boundary vector.
• GvecEnergy.J() returns the MHD energy at the solved equilibrium; dJ() returns the analytic reverse boundary adjoint gradient through adjoint_boundary_gradient, exact when the build provides Enzyme derivatives and FD otherwise.
• make_simsopt_optimizable(state) raises a typed ImportError without SIMSOPT; the problem class itself works without SIMSOPT.
• Tests skip SIMSOPT-dependent cases when SIMSOPT is missing and validate the adjoint gradient against centered boundary-perturbation finite differences with interior re-solves.

Closes #24.

Verification

Test fails on main

$ bash -lc 'test -f test-CI/unit-pygvec/test_ad_simsopt.py || { echo "missing SIMSOPT wrapper test before AD-23"; exit 2; }'
missing SIMSOPT wrapper test before AD-23

Test passes after fix

Enzyme-off isolated build (FD path):

$ pytest test-CI/unit-pygvec/test_ad_simsopt.py -q --require-pygvec
5 passed in 3.47s

Enzyme-on isolated build (exact path, fixed RZ, LLVM/Flang 22.1.6):

$ pytest test-CI/unit-pygvec/test_ad_simsopt.py -q --require-pygvec
5 passed in 1.18s

test_simsopt_wrapper_reuses_cached_equilibrium asserts one equilibrium solve for repeated dJ() and a new solve after a boundary change. test_simsopt_gradient_matches_fd_resolve checks sampled boundary DOFs against (J(x+δ)-J(x-δ))/2δ with full interior re-solves.