Backfill unit catalog + constants domains from main; fix Fahrenheit and Generic constants

[matt-edmondson]

Summary

First step of the parity plan: backfills the unit catalog and physical-constant domains that the vectors rework hadn't yet carried over from main, and fixes two real bugs found along the way.

Units (~60 across 30 dimensions)

• Imperial / US customary: nautical mile, knot, feet-per-second, acre, hectare, square mile/km/cm, cubic foot/inch/cm, imperial gallon, US quart/pint/fluid-ounce, stone, short ton, pound-force, BTU, Rankine, foot-candle
• CGS: dyne, erg, centipoise, dyne-per-centimeter
• Traditional radiological: curie, rad, rem, roentgen
• SI-prefixed conveniences: kPa, kN, kJ, kW, MW, kHz, MHz, mA, kA, kV, kΩ, MΩ, μF/nF/pF, kmol/mmol, ns, mcd, mrad
• Chemistry: millimolar, micromolar, dalton, enzyme unit, calorie-per-mole, ampere-hour
• Angle / ratio: gradian, revolution, ppm, ppb, percent-by-weight, standard gravity, torr, watt-hour, kilocalorie, atomic mass unit, g/cm³, g/L

Each unit gets a named conversion factor in conversions.json (exact values where defined), an explicit plural factoryName, and availableUnits wiring — From{Unit} factories and typed In(unit) support come out of the generator.

Constants

New domain constants in domains.json: Acoustics (reference sound pressure/intensity/power, Sabine constant), Optics (luminous efficacy), NuclearPhysics (atomic mass unit, nuclear magneton), FluidMechanics (standard air density, water surface tension).

Bug fixes

1. Fahrenheit conversion was inverted: the metadata carried the K→F transform (×1.8, −459.67), so Temperature.FromFahrenheit(32) computed −402 K and threw on the V0 non-negativity guard. Now ×5/9 +255.372 (exact). Nothing had tested it.
2. PhysicalConstants.Generic.*<T>() was entirely unusable: T.CreateChecked(PreciseNumber) throws NotSupportedException (PreciseNumber's TryConvertToChecked is unimplemented). The generator now emits PreciseNumber.To<T>(), the supported materialisation path. Also previously untested.

Deliberately not ported (with rationale)

• Luminance units (nit, cd/m², foot-lambert) — no Luminance dimension exists yet; belongs to the missing-quantities phase.
• PerSecond — no RateConstant dimension yet; same phase.
• MolesPerSecond — mislabelled on main (ReactionRate is volumetric, mol/(m³·s)).
• Main's Mathematical constant helpers (OneHalf, FourThirds…) — generated code uses literals.
• Dalton is defined as 1 g/mol on MolarMass rather than main's 1.66×10⁻²⁷ value, which was the per-particle mass, not a molar mass.

Testing

• 26 new tests in UnitBackfillTests covering factories across all the new unit families, In(unit) round-trips, the Fahrenheit/Rankine fix, and the new constants.
• Full suite: 664 pass (638 before + 26 new), 0 quantity regressions; the 93 failures are the pre-existing Windows-path tests that don't apply on a Linux runner.
• Build clean, no SEM001–SEM004 diagnostics.

https://claude.ai/code/session_01AUgbNJjDBDCwcbUfPK3pGW

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate failed

Failed conditions
1 Security Hotspot
0.0% Coverage on New Code (required ≥ 80%)
C Reliability Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE