instead of a max number of iterations, record terms that have been seen

instead of a max number of iterations, record terms that have been seen
in a hash instead.
Side effect: we no longer parse the same tc again and again, leading
to a smaller result with the same capabilities but no duplicates.
(instead of parsing 36 entries for xterm, we end up with 10 entries, as
the "tc=" stuff is actually a tree instead of a list: xterm references
several intermediate tc that all lead to the same basic list !)