The gain is not very impressive.  The runtime of the first example
in iface/dve2/README (also in dve2check.test) is 15% faster.

* src/tgba/tgbaproduct.hh (tgba_succ_iterator_product): Move ...
* src/tgba/tgbaproduct.cc (tgba_succ_iterator_product,
tgba_succ_iterator_product_common): ... in these two classes.
(tgba_succ_iterator_product_kripke): New class to speedup
successor computation on Kripke structures.  We can assume that
all the transitions leaving the same state have the same label.
(tgba_product::tgba_product, tgba_product::succ_iter): Use
tgba_succ_iterator_product_kripke when appropriate.
(tgba_product_init::tgba_product_init): Adjust for the case
where tgba_product did reverse its operands.