0.9.0: pin conductivity (LD 5) load behaviour and report out-of-range targets

[andrekuehne]

This PR adds the necpp side of WP0 for andrekuehne/PhasedArrayVisualizer-NG#159 (wire conductor material). The app computes conductor loss itself from NECQ currents, so it needs necpp's LD 5 behaviour pinned and documented exactly. It also bumps the WASM package to 0.9.0; after merge, release is the wasm-v0.9.0 tag on master.

Engine behaviour is unchanged. 0.8.0 already does everything the app relies on, including recomputing LD 5 on prepare() at a new frequency. This PR adds regression coverage, a normative contract, and one clearer error message.

Changes

• Tests: new src/nec_stateful_model_conductivity_tb.cpp (Catch2 [conductivity], runs under necpp_unit) and packages/necpp-wasm/test/conductivity-load.test.mjs. Together they pin:
  • structureLossW = Σ ½·Re Z′ᵢ·Lᵢ·|Aᵢ+Cᵢ|². This is the segment-centre rule with NEC ZINT, checked in all three branches (x ≤ 8, 8 < x ≤ 110, x > 110). It uses each segment's own radius, and feed segments are loaded too.
  • The app's GL-4 NECQ samples reproduce structureLossW to 1e-12 once each segment's exact centre current is rebuilt from the samples.
  • prepare(f1) → prepare(f2) is bitwise equal to a fresh prepare(f2). This holds for the stateful model and for NecArraySolver with symmetry off and auto.
  • With a perfect ground, loss counts physical segments only.
  • FORTRAN NEC-2 reference. testharness/FORTRAN/nec2dx.f was built with gfortran using the same c0. It agrees with necpp to every printed digit: Z to 6 significant figures, loss to 5. The decks are copper, stainless and σ = 1e3 dipoles and a copper monopole, each at 1000 and 1100 MHz.
  • The ring evaluator accepts loaded wires.
  • Validation happens at addLoad.
• Error message: a tag-relative load range that runs past the tag's last segment used to fail with "NO SEGMENT HAS AN ITAG OF 1", for a tag that exists. It now fails with "SEGMENT TARGET IS OUT OF RANGE FOR THE TAG". The error class (NecGeometryError) and the absent-tag message are unchanged.
• Docs: docs/wasm-api.md gains a normative "Conductivity loads (LD 5)" section: the formula and its branches, NEC's constants, the accounting rule, recompute on frequency change, and validation.
• Version: 0.9.0 in the same places the 0.8.0 bump touched: package.json, the lockfile, versions.ts, the fixture manifest, the README CDN example, and the two version-asserting tests.
• CHANGELOG: new 0.9.0 section. The GE -1 entries that sat under "Unreleased" are in the commit tagged wasm-v0.8.0, so they now sit under 0.8.0.

Note for anyone using nec2c as a golden

nec2c 1.3.1 does not match NEC-2 for x ≤ 110. On the same decks it is 2 % off for stainless and 19 % off for σ = 1e3. necpp, the original FORTRAN and the exact J0/J1 form all agree with each other.

nec2c agrees only in the x > 110 branch (copper at x ≈ 203: 1 GHz, a = 0.3 mm). The visualizer's product metals at 10 GHz (copper x ≈ 64) all sit in the series branch, where nec2c is wrong.

Validation

Run locally with the Emscripten SDK 4.0.7 build of this branch:

• native ctest: 8/8 pass;
• npm --prefix packages/necpp-wasm test: 166/166 pass, typecheck clean;
• pack manifest test passes at 0.9.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LEzEibjCJrRrYxMm1uke1N