Created DocComment class that strips all the asterisks from
doc-comment, leaving only the real contents.

Completely rewrote the lexer, so that the whole
input is tokenized right away.  This enables us
to easily look ahead for several tokens.