<bit>: countr_zero() doesn't need to special-case 0 for uint8_t and uint16_t

[StephanTLavavej]

std::countr_zero and std::countr_one in <bit> are powered by the following machinery:

STL/stl/inc/bit

Lines 249 to 257 in 335449f

	template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> = 0>
	_NODISCARD constexpr int countr_zero(const _Ty _Val) noexcept {
	return _Countr_zero(_Val);
	}
	template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> _Enabled = 0>
	_NODISCARD constexpr int countr_one(const _Ty _Val) noexcept {
	return _Countr_zero(static_cast<_Ty>(~_Val));
	}

We want to use the tzcnt instruction, but if the processor doesn't support it, it'll be treated as bsf which doesn't handle 0 the way we want, so we need to special-case it. We have logic to avoid the special-casing when we detect that tzcnt is available at compile-time or run-time:

STL/stl/inc/limits

Lines 1055 to 1065 in 335449f

	template <class _Ty>
	_NODISCARD int _Checked_x86_x64_countr_zero(const _Ty _Val) noexcept {
	constexpr int _Digits = numeric_limits<_Ty>::digits;
	constexpr _Ty _Max = (numeric_limits<_Ty>::max)();
	#ifndef __AVX2__
	const bool _Definitely_have_tzcnt = __isa_available >= __ISA_AVAILABLE_AVX2;
	if (!_Definitely_have_tzcnt && _Val == 0) {
	return _Digits;
	}
	#endif // __AVX2__

Charles Milette (@sylveon) noticed that we can optimize this further. This is immediately followed by:

STL/stl/inc/limits

Lines 1067 to 1072 in 335449f

	if _CONSTEXPR_IF (_Digits <= 32) {
	// Intended widening to int. This operation means that a narrow 0 will widen
	// to 0xFFFF....FFFF0... instead of 0. We need this to avoid counting all the zeros
	// of the wider type.
	return static_cast<int>(_TZCNT_U32(static_cast<unsigned int>(~_Max | _Val)));
	} else {

For uint8_t and uint16_t, this 1-bit-filled widening means that we'll never actually pass 0 to the instruction, so the ISA check is entirely unnecessary! We can reorganize this code accordingly.