To expand on the comments, both your version and the version of @Andreas compile to the same asm (using clang):

https://godbolt.org/z/qoq9ejEoa

auto subtract26(std::uint8_t x, std::uint8_t y) -> std::pair<std::uint8_t, bool>
{
    std::uint8_t result;
    bool overflow = ckd_sub( &result, x, y );

    return { result, overflow };
}

auto subtract(std::uint8_t x, std::uint8_t y) -> std::pair<std::uint8_t, bool>
{
    return {x - y, x < y};
}

subtract26(unsigned char, unsigned char):
        xor     ecx, ecx
        sub     dil, sil
        setb    cl
        shl     ecx, 8
        movzx   eax, dil
        or      eax, ecx
        ret

subtract(unsigned char, unsigned char):
        xor     ecx, ecx
        sub     dil, sil
        setb    cl
        shl     ecx, 8
        movzx   eax, dil
        or      eax, ecx
        ret

So even if you do not have access to C++26, your compiler should be smart enough to optimize this :)