Instead of recursive calls for each nested level of HTML, the next level is
pushed to a stack on the heap, and returned to the main loop. This prevents
stack overflow at _n_ depth (where _n_ is operating-system dependent). It's
probably still possible to use all of the heap memory, but Tidy's allocators
already fail gracefully in this circumstance.

Please report any regressions of your own HTML!

NOTE: the XML parser is not affected, and is probably still highly recursive.