Someone (I need to refer back) asked that we do the same skip for the
  utf-8 win32 prefix when testing for shebang lines.  Here's just such
  a test with some bad sizeof(buffer) v.s. bytes read assumptions fixed,
  and the code made generally a little more legible.  Please review and
  comment to consider this patch for backporting to 2.0.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@98775 13f79535-47bb-0310-9956-ffa450edef68