Generic degree-rendering formatters, millifahrenheit fix, drop *_real types (2.0.0)

[cleishm]

Summary

• Generic formatters: the 18 hand-written std::formatter specializations and 15 to_string overloads are replaced by two generic templates each. An empty format spec keeps the exact-count form (millicelsius(22500) → "22500m°C"); any floating-point spec renders the value in scale degrees, so readable display no longer needs a lossy cast:

  std::format("{:.1f}", millicelsius(22534));  // "22.5°C"
  std::format("{:.1f}", delta_millifahrenheit(9000));  // "9.0Δ°F"

  Width/alignment pass through to the numeric part; specs on integer types were compile errors before, so this path is purely additive.

• ⚠️ BREAKING — millifahrenheit ratio fix: millifahrenheit and delta_millifahrenheit used std::ratio<5, 900> (0.01 °F per count) instead of the documented std::ratio<5, 9000> (0.001 °F). Counts change meaning by 10×: temperature_cast<millifahrenheit>(fahrenheit(1)).count() was 100, is now 1000. The new conversion tests were written first and confirmed to fail against the old ratio.

• ⚠️ BREAKING — *_real typedefs removed: celsius_real, kelvin_real, fahrenheit_real existed mainly as a display workaround; spec-based rendering covers that directly. Generic floating-point rep support in the templates remains (e.g. temperature<celsius_scale, delta<double>>). Migration: replace std::format("{:.1f}", temperature_cast<celsius_real>(x)) with std::format("{:.1f}", x).

• Version bumped to 2.0.0 per semver (removed public API + count reinterpretation).

Test plan

• New conversion tests demonstrably failed before the ratio fix, pass after
• All nine empty-spec output strings pinned byte-for-byte (backward compat)
• New spec-rendering test matrix across scales, precisions, deltas, width pass-through
• Non-format build path verified: -DCONFIG_THERMO_STD_FORMAT=0 -fsyntax-only
• 29/29 tests pass