Physical coefficients in simsopt/DiscretizedCoils conversions; array PerturbationSample - #139

Curves.dofs (and Coils.dofs_curves) are the optimizer-facing coefficients, _dofs * scaling; the geometry is built from the unscaled _dofs. Three conversions used the scaled values as physical Fourier coefficients, so any coil set with non-default scaling_factor/scale_fixed was exported or converted with the wrong shape:

• Curves.to_simsopt and Coils.to_simsopt now use _dofs (the coil export also takes dofs_currents_raw directly instead of dofs_currents * currents_scale).
• DiscretizedCoils.from_Coils now builds the base gamma from coils.curves._dofs. On main, for a radius-2 circle with scaling_factor=0.3, scale_fixed=7, the converted gamma differs from coils.gamma by up to 16.9 m.

Also PerturbationSample.__init__ used if sample: on a JAX array, which raises for any multi-element sample; it is now if sample is not None:.

Tests (fail on main, pass here; JAX_ENABLE_X64=1 pytest -n 2 tests/test_coils.py tests/test_coil_perturbation.py):

• test_scaled_curves_convert_with_physical_coefficients uses a stub simsopt module so it runs in CI without simsopt, and checks both exports and from_Coils against the physical coefficients and gamma.
• test_perturbation_sample_accepts_array_sample

On main: 2 failed. With this branch: 37 passed.

Overlap: #132 fixes the related DiscretizedCoils.copy / perturb_curves use of the scaled dofs_gamma and does not touch these lines. This PR does not duplicate it and merges cleanly with #132, #119 and #133.