Instead of deriving different tokens from a base Token class, we now use an enumerated TOKEN_TYPE to distinguish types. This is so we don't have to cast all the time when parsing the resulting token stream.
Also, removed start/end stream tokens.