[RFC] Fix expect leaky pipes

[Christopher Faylor]

Ok.  I've redone the pty master stuff to use archetypes, in a manner so
similar to the slave that I almost think there could be some commonality
extracted into tty_common.  However, I'm not going to do that yet.

It seems to work but it is a big change, although I think there probably
is a net loss in code size.

I'm going to set an appropriate pre-CVS tag and then check something in.
I'd appreciate if you all could check this before I release a snapshot.
I've tried my changs with CYGWIN=tty, rxvt, xterm, and openssh but there
are bound to be problems given the pervasiveness of this patch.

As I mentioned in cygwin-patches, after Dave's investigation led me to
think about this, it seems like some of this stuff worked only by
coincidence and relied on predictable ways of using pty masters.  I hope
this patch doesn't have that particular flaw.

The one problem that it does have is that it doesn't deal transparently
with close-on-exec of pty handles with non-cygwin applications.  That
was already a problem with the slave part and it is something that can
be fixed eventually if it turns out to really present a problem.

cgf