- fix recognition of newline in INITIAL state; this exposes a problem
  where the lexer matches unknown tokens for too long at the end of file
- fix recognition of unknown tokens at the end of file
- call `yyterminate()` at the end of file