Method sqrt

cpp/fory/util/float16.h:169–171

  // ---- Optional math (computed in float32, rounded back) ----

  [[nodiscard]] static float16_t sqrt(float16_t a) noexcept {
    return from_float(std::sqrt(a.to_float()));
  }
  // fmin/fmax propagate NaN the same way as IEEE minNum/maxNum.
  [[nodiscard]] static float16_t min(float16_t a, float16_t b) noexcept {
    return from_float(std::fmin(a.to_float(), b.to_float()));