Instead of opening the file in unicode mode, we need only encode
data which potentially has non-ASCII characters as UTF8 before
writing. This should work across both Python versions, and is
also far simpler than anything else discussed.
llvm-svn: 258969