---------------------------------IDEAS----------------------------------

1. Have a more general means of representing paragraph transformations
   and applying them so that text2html can be extended.  Each
   transformation would have (a) an option name, (b) a sub for testing
   whether the transformation is applicable to the paragraph and (c) a
   sub for applying the option.  (There must be a means of passing
   information from step (b) to step (c) as for the table
   transformation.)

   (But are there that many other transformations that can be applied on
   a paragraph basis?  Other kinds of transformations would be harder to
   generalise like this.  Think about the need to have a different idea
   about where paragraph boundaries are depending on the options.)

2. The current approach is to test the transformations for applicability
   one by one and apply the first one that matches.  It might be better
   to test all the transformations and score them, then apply the
   best scoring.  This would allow heuristics to be applied.

3. The current approach goes through the paragraphs one by one with no
   lookahead.  Multiple passes through the paragraphs could allow
   heuristics to be applied, e.g., to distinguish nested lists from
   adjacent lists.

4. These transformations could be implemented:

   * lettered paragraphs a) b) c) etc

   * underlined headings

   * quoted text in e-mail message (impossible to accurately given the
     enormous diversity of quoting styles -- would 60% be good enough?)