In optimized x64 code, the check for minus zero after multiplication
used the wrong size bit-or instruction (64-bit instead of 32-bit).

TEST=mjsunit/numops-fuzz.js
Review URL: https://chromiumcodereview.appspot.com/9316127

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@10606 ce2b1a6d-e550-0410-aec6-3dcde31c8c00