20000131 anomolous system header warnings

[Bernd Schmidt]

> We have to keep the obstack stuff indefinitely because of languages that are
> not integrated.  For the same reason TREE_PERMANENT is not an obsolete concept.

Still, any place that uses TREE_PERMANENT (or any other means of determining
which obstack a given node is on) is broken if we want to remove obstack
code from frontends that have been converted to gc.  Eliminating such uses
is a good idea.

Bernd