Revert "[libc] Refactor BigInt" (llvm#87612)

Reverts llvm#86137

Some aarch64 compilers seem to consider that `uint128_t` is not
`is_trivially_constructible` which prevents `bit_cast`-ing.

Author: gchatelet

libc/fuzzing/CMakeLists.txt

@@ -1,7 +1,6 @@
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=fuzzer")
 add_custom_target(libc-fuzzer)
 
-add_subdirectory(__support)
 # TODO(#85680): Re-enable math fuzzing after headers are sorted out
 # add_subdirectory(math)
 add_subdirectory(stdlib)

libc/fuzzing/__support/CMakeLists.txt

@@ -58,9 +58,9 @@ template <size_t Bits> struct DyadicFloat {
   // significant bit.
   LIBC_INLINE constexpr DyadicFloat &normalize() {
     if (!mantissa.is_zero()) {
-      int shift_length = cpp::countl_zero(mantissa);
+      int shift_length = static_cast<int>(mantissa.clz());
       exponent -= shift_length;
-      mantissa <<= static_cast<size_t>(shift_length);
+      mantissa.shift_left(static_cast<size_t>(shift_length));
     }
     return *this;
   }
@@ -233,7 +233,7 @@ LIBC_INLINE constexpr DyadicFloat<Bits> quick_add(DyadicFloat<Bits> a,
     result.sign = a.sign;
     result.exponent = a.exponent;
     result.mantissa = a.mantissa;
-    if (result.mantissa.add_overflow(b.mantissa)) {
+    if (result.mantissa.add(b.mantissa)) {
       // Mantissa addition overflow.
       result.shift_right(1);
       result.mantissa.val[DyadicFloat<Bits>::MantissaType::WORD_COUNT - 1] |=