v3.0.0

Added

• Property-based tests (t_property) checking algebraic identities such as
  (a * b) / b == a and inv(inv(x)) == x across randomized values, as
  regression insurance for the mapper and operators.
• README status badges (CI, docs, latest release, license, C++17).
• Release automation: pushing a v* tag publishes a GitHub Release with notes
  drawn from the matching CHANGELOG.md section
  (tools/extract_changelog.sh, .github/workflows/release.yml).
• Zero-overhead proof (tools/check_zero_overhead.py, run in CI for gcc and
  clang): compiles unit arithmetic and the equivalent raw arithmetic to
  assembly at -O2 and asserts the two are byte-for-byte identical, making
  the zero-overhead claim a regression test.
• .clang-format codifying the house style, with a CI format job that fails
  on any unformatted file. clang-format is pinned so local and CI agree.
• Mixed-precision arithmetic: same-dimension +, -, the comparison operators
  and the cross-unit * / / now accept operands with different storage types
  (e.g. Length<float> + Length<double>) and promote the result to the wider
  type via std::common_type_t.
• TemperatureDelta type for affine-correct temperature arithmetic: a
  difference of temperatures is a displacement, not an absolute point, so its
  degC() / degF() apply only the scale factor and never the +273.15 /
  +32 offset. Supports Delta ± Delta, scalar * / /, unary -, and
  Temperature ± Delta.
• Packaging: a Conan recipe (conanfile.py with a test_package) and a vcpkg
  overlay port (packaging/vcpkg/ports), plus README install instructions for
  Conan, vcpkg and CMake FetchContent. The imported target is utl::utl
  everywhere.
• Angle as an eighth pseudo-dimension: Angle (rad) and SolidAngle (sr) are
  now distinct dimensions (rad = angle¹, sr = angle²), so Angle * Angle == SolidAngle, sqrt(SolidAngle) == Angle, and LuminousIntensity * SolidAngle == LuminousFlux. A new Torque (N·m = energy/angle) is type-distinct from
  Energy (Energy / Angle == Torque, Torque * Angle == Energy). Added
  sin / cos / tan overloads that take an Angle and return a scalar.

Changed (breaking)

• BaseUnit gained an eighth template exponent ANGLE (defaulted to 0, so
  named-unit code is unaffected) and dim() now returns std::array<int8_t, 8>.
  Angle and SolidAngle are no longer dimensionless — they carry the angle
  dimension — and LuminousFlux (cd·sr) is now type-distinct from
  LuminousIntensity (cd). Code reading dim() as a 7-element array, or
  relying on angles comparing equal to raw scalars, must adjust.

• ThermodynamicTemperature - ThermodynamicTemperature now returns a
  TemperatureDelta (was ThermodynamicTemperature), and
  ThermodynamicTemperature + ThermodynamicTemperature is deleted: adding two
  absolute temperatures is physically meaningless. Add a TemperatureDelta
  instead (temp + TemperatureDelta(5)). Reading a temperature difference
  via .degC() / .degF() previously misapplied the scale offset.

• BaseUnit default constructor is deleted: a physical quantity is never
  undefined. Every unit must be constructed with an explicit value. Callers
  that relied on default construction (e.g. std::map::operator[],
  value-initialized std::array) must initialize explicitly.