perf(math): scan least significand bit with MSVC x64 intrinsic

src/fesa/math/high_low_arithmetic.h

@@ -7,6 +7,9 @@
 #include <cstring>
 #include <initializer_list>
 #include <limits>
+#if defined(_MSC_VER) && defined(_M_X64)
+#include <intrin.h>
+#endif
 #if defined(_M_X64) || defined(__SSE2__)
 #include <xmmintrin.h>
 #endif
@@ -114,8 +117,14 @@ inline int LowestBitExponent(double value) {
                            ? -1074
                            : static_cast<int>(encoded_exponent) - 1075;
   if (encoded_exponent != 0U) significand |= 1ULL << 52U;
+#if defined(_MSC_VER) && defined(_M_X64)
+  // The finite, nonzero precondition makes the significand nonzero.
+  unsigned long trailing = 0;
+  _BitScanForward64(&trailing, significand);
+  return exponent + static_cast<int>(trailing);
+#else
   int trailing = 0;
-  // Strip large zero groups first without vendor/compiler intrinsics.
+  // Preserve the portable search outside the supported MSVC x64 path.
   for (const int shift : {32, 16, 8, 4, 2, 1}) {
     const auto mask = (1ULL << shift) - 1U;
     if ((significand & mask) == 0U) {
@@ -124,6 +133,7 @@ inline int LowestBitExponent(double value) {
     }
   }
   return exponent + trailing;
+#endif
 }
 /// @brief Gives a product and its FMA residual when the dyadic tail fits.
 /// @note Rejects underflow-risk exact tails, even if the rounded high is finite.