Enforce up-down symmetry throughout inverse solves

[nicamo]

Summary

• Project the initial plasma flux onto its up-down-even component before the first inverse residual and current update when force_up_down_symmetric=True.
• Propagate symmetry forcing into the full plasma-response Jacobian builder.
• Apply it to both the baseline equilibrium and every finite-difference coil perturbation solve.
• Preserve the existing default behaviour and positional compatibility of optimize_currents().
• Clarify the symmetry contract in Example 07a and report the resulting plasma-flux parity error.

Motivation

Previously, symmetry forcing was passed only to the main forward corrections performed by the inverse loop. The initial inverse update could therefore react to an odd component in the supplied plasma flux, while the baseline and perturbed forward solves used to construct the full Jacobian silently ran without symmetry forcing. This meant the inverse solve did not consistently remain in the requested even-plasma subspace.

The flag constrains the plasma response; exactly symmetric currents in independent upper/lower coils still require those currents to be linked or constrained by the machine/control model.

Validation

• Added coverage proving an explicitly asymmetric initial plasma_psi is projected before inverse iteration.
• Added coverage proving the full-Jacobian baseline and every perturbed solve receive the symmetry flag.
• Existing diverted inverse regression remains unchanged.
• test_inverse_static_solver.py and test_static_solver.py: 4 passed.
• Example 07a runs end-to-end.
• Its first symmetric Anamak inverse case converges in 9 iterations, exercises the full-Jacobian path, reaches 4.39e-6, and reports a relative plasma-flux parity error of 0.000e+00.

Scope

The runtime implementation adds only the initial projection and missing argument propagation. Behaviour is unchanged when force_up_down_symmetric=False. The example amendment adds one explanation and one parity diagnostic.

[kpentland]

Sensible fix!