- Give more information in callbacks, such as when the beginning and
  end braces of a tag occurred (offset in the stream)

- Provide a way to parse entity-refs.  This will be handy for usertrack
  in that we can then just specify a &USERTRACK; entity anywhere, and
  have it substituted with the session ID.

- Make sure comments and <!FOO tags are processed directly.  Provide the
  option of either passing them through as raw data or as actual structures

- Possibly make the return value of a flushing operation more usable by
  external programs.  I.e. usertrack's callbacks simply pass the brigade,
  and really have no way of returning any result information back up the
  stack and out of a feed or flush operation.

- Perhaps speed up some of the charmap lookup operations by making a 64k
  table, and comparing short values.

- Put some goto's in the starttag parsing to immediately jump between
  some of the state transitions.

- Maybe make starttag registering return an integer, representing the
  tag that is requested.  That way callbacks won't have to strcmp, and
  can just use the integer.