Fix formatting regression introduced by r174307.

In preprocessor definitions, we would not parse all the tokens and thus
not annotate them anymore. This led to a wrong formatting of comments
in google style:

  #endif // HEADER_GUARD  -- requires two spaces

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@174361 91177308-0d34-0410-b5e6-96231b3b80d8
2 files changed