Use ldexp instead of pow for more accuracy

src/conversions-inl.h

Index: src/conversions-inl.h
===================================================================
--- src/conversions-inl.h	(revision 11248)
+++ src/conversions-inl.h	(working copy)
@@ -230,7 +230,7 @@
   ASSERT(number != 0);
   // The double could be constructed faster from number (mantissa), exponent
   // and sign. Assuming it's a rare case more simple code is used.
-  return static_cast<double>(negative ? -number : number) * pow(2.0, exponent);
+  return ldexp(static_cast<double>(negative ? -number : number), exponent);
 }