[minor] Convert units exactly for every storage type

[matt-edmondson]

Summary

Unit conversions and vector lengths are now exact to the precision of the storage type, not just to the precision of double. With decimal storage, every conversion factor is correct to 28 digits and Length() returns √2 as 1.4142135623730950488016887242. double and float behave as before, apart from two factors that moved one step closer to their exact values.

This is also what a non-IEEE storage type such as PreciseNumber needs. With PreciseNumber 2.0 as storage, Length.FromFoot(1).In(Units.Inch) is exactly 12, where double gives 12.000000000000002.

Changes

• Factors for each storage type. ConversionsGenerator and MagnitudesGenerator emit a nested Values<T> holder, parsed once for each closed generic type, next to the existing const double. Generated From{Unit} factories, IUnit.ToBase<T>, FromBase<T>, and every In(unit) read it. Integer storage, and any type that can't parse the literal, falls back to T.CreateChecked(double).
• Exact metadata. conversions.json accepts a fraction such as "5/18". A malformed value reports the new SEM009 error. Thirteen factors that were stored already rounded now carry their exact definitions: repeating ratios as fractions, and π-based factors as 150-digit literals. SquareMile, PoundForce, and PoundFoot were already exact and are unchanged.
• Square roots. Every vector's Length() and Distance() call the internal StorageMath.Sqrt. double, float, Half, and the integer primitives keep their previous expression exactly. Other types refine the double estimate with Newton steps until the value stops changing.
• Public additions. IUnit.ToBaseFactorAs<T>() and ToBaseOffsetAs<T>() have default implementations, so existing implementers keep compiling. StorageTypes now lists decimal.
• Docs. CLAUDE.md and docs/physics-generator.md describe the fraction syntax, the per-type factors, and SEM009.

Risk

• 172 generated files changed. The only removed lines are the intended rewrites: T.CreateChecked(constant) becoming Values<T>, the double square root becoming StorageMath.Sqrt, the 13 metadata constants, and the StorageTypes list.
• PsiToPascals is a long literal rather than a fraction. As a fraction, float storage would land further from the exact value, because the numerator would have to round to a float before the division.
• An older mismatch remains unchanged. QuantitiesGenerator ignores conversionFactor when a unit also has a magnitude, and UnitsGenerator applies both. No unit has both today.

Testing

• 1227 tests pass, 0 fail, and 4 are skipped (the g++ and clang++ compile tests, since no compiler is installed locally). 59 of the tests are new.
• StorageConversionTests<T> runs the same conversion and relationship checks for double and decimal, including exact decimal results such as a foot being 12 inches and a 3-4-0 vector having length 5. Adding a storage type takes one derived class.
• UnitFactorConsistencyTests checks that every unit's ToBaseFactorAs<double>() matches its ToBaseFactor property.
• The existing zero-allocation tests still pass, and there's a new one for In(unit).
• Generate-AliasProps.ps1 produces no content changes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01K5Bk9UjGdGUtC5C6qK5ZxD

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
88.2% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud