feat: complete Phase 3 roadmap + COST 231 + ITM variability modes

[d3mocide]

Knocks out the remaining roadmap items ahead of the next production build. Closes P3-1, P3-3, P3-4, P4-2 and P4-6.

Important

This PR rebuilds meshrf.wasm and regenerates libmeshrf/js/meshrf.js — the module loader every WASM hook imports. See Verification of the WASM rebuild below for what I checked and what still needs a human pass before shipping.

P3-1 — Client-side Hata / FSPL

Okumura-Hata is now implemented in JavaScript (src/utils/math/hata.js), mirroring rf_physics.calculate_hata_loss term-for-term including the input clamping. Both engines agree to 4 decimal places on the reference cases (900 MHz / 5 km / 50 m / 1.5 m → 146.9428 dB on each side).

A new dispatcher (src/utils/math/pathLoss.js) lets LinkLayer resolve fspl and hata locally, so neither model needs /api/calculate-link and both keep working offline / in the PWA.

bullington and server-side itm deliberately stay on the backend — the Python implementation applies clutter_height and k_factor to the terrain profile, and the current JS calculateBullingtonDiffraction does not, so moving them would silently change existing results.

P4-2 — COST 231-Hata extension

Extends Hata coverage from 150–1500 MHz to 150–2000 MHz in both engines. Selecting "Hata" auto-dispatches to COST 231 at or above 1500 MHz; cost231 is also accepted as an explicit backend model.

Per the standard, COST 231 defines only the 3 dB metropolitan correction and has no suburban or rural term. Rather than silently reusing the Okumura-Hata corrections at frequencies where they aren't defined, selecting those environments above the crossover now raises an explicit warning. Link Analysis validity warnings are driven from a single getHataValidity helper instead of inline thresholds.

P3-3 — WASM ITM for batch reports

Batch Processing gained a propagation-model selector:

• Bullington (Fast) — FSPL + knife-edge diffraction over 20-point profiles (unchanged default).
• Longley-Rice ITM — the same WASM engine Link Analysis uses, over a 100-point profile, honouring Ground Type, Climate Zone and Reliability, so batch numbers now match single-link analysis.

The WASM module loads lazily on selection — useWasmITM gained an enabled flag so this always-mounted panel doesn't pull a ~MB instantiation onto the startup path. A link that fails ITM falls back to Bullington rather than failing the whole report, and the Model column records which model actually produced each row.

P3-4 — Per-node configs in batch CSV

CSV import now reads optional Antenna_Height, Antenna_Gain, TX_Power, Device and Antenna columns, with aliases (Height/AGL/Gain/Power) and loose preset matching — HELTEC_V3, heltec-v3 and Heltec V3 all resolve to the same device. An Antenna preset supplies its nominal gain; an explicit Antenna_Gain wins.

nodeConfig.js merges overrides over the global A/B config field-by-field. Blank cells and existing three-column files behave exactly as before, so this is backward compatible.

P4-6 — ITM statistical variability modes

Previously time_pct was hardcoded to 50.0 in meshrf_itm.cpp, so every prediction was the median forecast with no way to plan for worst-case conditions.

C++ / WASM: added time_pct, loc_pct, sit_pct and mdvar to LinkParameters, exposed them through Embind, and threaded variability into the coverage path — calculate_rf_coverage gained three trailing arguments, so useRFCoverageTool now passes 20 (Embind matches on arity).

Frontend: a Reliability selector in the Environment sidebar offers Best Case (10/10/10), Typical (50/50/50, default) and Reliable (90/90/90), wired through Link Analysis, RF Coverage (click, recalc and drag paths) and batch ITM reports. Batch rows record the mode that produced them (e.g. ITM 90/90/90).

Measured on a synthetic ridge (915 MHz, 10 m TX / 2 m RX) — roughly a 20 dB spread:

Mode	Predicted loss
Best Case 10/10/10	193.49 dB
Typical 50/50/50	203.76 dB
Reliable 90/90/90	213.66 dB

Verification of the WASM rebuild

public/meshrf.wasm, libmeshrf/js/meshrf.wasm and libmeshrf/js/meshrf.js are regenerated artifacts. Built with emcc 6.0.4 via the repo's own Dockerfile.wasm recipe; ROADMAP P4-6 documents the exact command.

Checked against the installed artifacts, not assumed:

• All four bindings still present (calculate_itm, calculate_rf_coverage, calculate_viewshed, LinkParameters).
• LinkParameters defaults intact for every pre-existing field, plus the four new ones.
• The new 20-argument calculate_rf_coverage call succeeds and returns a full grid.
• calculate_viewshed still works — it shares the rebuilt glue even though this change doesn't touch it.
• Backward compatibility: a call that never sets the new fields returns a bit-identical result to an explicit 50/50/50.

What this does not cover: emcc 6.0.4 is almost certainly not the version that produced the previously committed artifact (the JS glue moved 48,650 → 48,130 bytes). Unit tests can't catch a loader regression, so link analysis, RF coverage and viewshed each want a manual pass in the running app before this ships to prod.

Also in this PR

• Bug fix: Bullington diffraction was never displayed for the Hata model in LinkLayer — the check compared against 'Hata' while the model value is lowercase 'hata', so the condition could never be true.
• Mesh report export writes via a Blob instead of a data: URI and escapes CSV fields, so node names containing commas or quotes no longer corrupt the report.
• Mesh report gained Model, PathLoss_dB and per-node TX/RX height/gain/power columns, plus a progress bar.

Still open

P3-2 (model selection for RF coverage) — needs a JS raster path over the stitched elevation grid, now easier since calculateClientPathLoss exists.

P4-6 statistical contours — the variability controls are done; rendering probability-of-reception contours from multiple percentage runs is tracked as the remaining piece.

Testing

• Frontend: 52 passed (36 new — Hata/COST 231 math against hand-computed references, the client-side dispatcher, CSV override parsing, config merging, and the variability mapping).
• Backend: 23 passed (14 new in rf-engine/tests/test_rf_physics.py, pinning the same reference values so the two engines can't drift apart).
• npm run lint and npm run build clean.

Docs updated: ROADMAP.md, CHANGELOG.md, README.md, Documentation/batch-processing.md, Documentation/link-analyzer.md.