all-backend ledger: xfail jax test_harmonic_planar_stm_vs_fd_of_flow

Restores the jax entry for this borderline FD-of-flow STM check that the
#969 milestone-1 regen-prune erroneously dropped (the jax SpiralArms/
potential/scf shard truncated mid-run, issue #53). Under jax the Multipole
(non-axi) coefficients differ from numpy at the coefficient level, pushing
this ~8e-4 FD-of-flow-vs-C-STM comparison just over the 5e-4 tolerance;
numpy passes (verified), the C STM is correct. Not a #972 regression — this
PR only adds C-STM autodiff files. Tolerance unchanged.

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: jobovy

tests/backend_xfail.txt

@@ -121,6 +121,7 @@ jax tests/test_potential.py::test_TimeDependentAmplitudeWrapperPotential_against
 jax tests/test_potential.py::test_amp_mult_divide
 jax tests/test_potential.py::test_evaluateAndDerivs_potential
 jax tests/test_potential.py::test_forceAsDeriv_potential
+jax tests/test_potential.py::test_harmonic_planar_stm_vs_fd_of_flow
 jax tests/test_potential.py::test_poisson_potential
 jax tests/test_potential.py::test_poisson_surfdens_potential
 jax tests/test_potential.py::test_potential_at_infinity