Consistently use UTF-8 instead of platform default charset

Fixes platform-dependent behavior when using non-ASCII characters in
grammar source files.