Write dimensional helical response output

[krystophny]

Closes #141.

Parent stack head: feat/flux-pumping/local-helical-drive at 47343f3 (PR #140). This cumulative PR targets main and remains unmerged with the rest of the feature stack.

Write helical_current_response.h5 for single-species mode-1 helical-source runs. The contract records the input quadrature, Fourier phase, mode, surface, species, solver switches, source amplitudes, normalized current channels, dimensional D31/D32 coefficients, and whether a separate off reference is required. It does not label a flux-surface moment as local or geometric toroidal current.

Correct the existing multispecies HDF5 unit metadata to match the normative NEO-2 output document: D13/D23/D31/D32 use cm^2 G/s; D33 uses cm^2 G^2/s.

Numerical and physics invariants:

• D_3k = gamma_3k v_T rho B_ref = gamma_3k 2 T c/(z e).
• Solver order and final-output mapping remain current row 3, A1/A2 columns 1/2.
• Fourier phase remains exp(+i(m_theta theta + m_phi phi)).
• The real/imaginary reconstruction is H=(R-R0)-i(I-R0); source-only fileless runs use the existing exact-zero guard as R0=0.
• The kinetic operator, collision moments, discretization, convergence criteria, transfer algebra, ABI, and memory layout are unchanged.
• The output is a harmonic response coefficient. Pointwise parallel current and geometric toroidal current require a later distribution/coordinate reconstruction.

Verification

Failing before:

helical_current_response.h5: absent
D31/D32 HDF5 unit attribute: cm^2/s

Passing after:

$ fo
Static: OK (340 modules, 340 changed, 340 affected)
Build: OK
Tests: OK
Lint: OK
All stages passed

The focused normalization/reconstruction, shared-source, and output-geometry tests pass 3/3. A three-case HELCORE integration completed every 24 x 24 final join in 3:24.90 with peak RSS 541996 KiB and no IEEE, NaN, signal, or runtime marker. The runner validated every HDF5 field, unit, phase, switch, species, finite value, and the agreement with fulltransp at relative tolerance 1e-13.

Reconstructed (m_theta,m_phi)=(1,-1) response:

D31 = 1.4677305578740989e12 - 5.42208046219255e11 i cm^2 G/s
D32 = 4.18887130153984e12 - 6.277850811343069e11 i cm^2 G/s

The exact-zero source guard supplies the baseline for this fileless source-only case.

[krystophny]

Added slice: pointwise helical parallel-current profile (commit 6439ac9)

Extends the resolved-current output with a per-field-line-point current profile
for the mode-1 local (axi) solver.

What it does. After the transport contraction
qflux(2,k) = sum_col flux_vector(2,col)*source_vector(col,k), the new
pointwise_current_mod decomposes that current channel into its per-point
contributions. flux_vector's columns partition into disjoint per-step blocks,
so the block dot products sum back to qflux(2,k) exactly. The routine reuses
the already-verified current row: the counter-passing block carries the explicit
minus sign (parallel current is odd under sigma -> -sigma) and the phi-quadrature
weight is baked in, so no new sign, weight, or normalization is introduced.

Invariant preserved. The solver writes current_profile.dat only when the
helical drive is active (isw_hel_drive /= 0), so runs without it are bitwise
unchanged. A hard self-check aborts the run if the field-line sum does not
reproduce qflux(2,k).

Verification.

• ctest TEST suite: 17/17 pass; new pointwise_current_test added.
• Red/green: breaking the block partition (dropping the npassing+1 factor)
  aborts with ERROR STOP FAIL: per-point contributions do not sum to the qflux current channel; restored -> Passed.
• Single-surface l=1 helical-core benchmark, three drive quadratures: the
  harmonic current response reproduces the reference to ~1e-11, and each
  current_profile.dat reconstructs qflux(2,k) to ~1e-15 relative over 961
  finite points.

Parent slice: dimensional helical response contract (this PR's earlier commits).
No merge requested; review only under the integration freeze.

[krystophny]

Correction: the pointwise current-profile slice was moved to its own stacked PR #151 (top of the stack, on the source-conjugacy branch) to keep this PR focused on the dimensional response contract and to avoid rebasing the intervening open PRs. Please disregard the previous comment's commit reference; the pointwise work is reviewed in #151.