RFC: specdb refactor for later interface<->library binding

[Jeff Johnson]

This is filed to promote some discussion.

We now have several cases of interfaces which exist in several libraries in
essentially the same form. There are a handful in libc/libpthread, there are
many shared between libpng12/libpng15, also many between
libncurses/libncursesw, and there will presumably be lots between
libgtk-2.0/libgtk-3.0.

All details of those interfaces have to be duplicated, which is irritating and
potenially error prone, because the Interface.Ilibrary field binds an interface
specifically to a library. So what if we changed to make that binding in the
ArchInt entry instead? That way, in theory at least, a common interface
definition could be bound to two or more separate libraries.

There's certainly plenty of rework that would be needed to support such a
change so I'm not proposing it as a high priority, just starting the
discussion.

Possible similar case is for Constant.Cheadgroup, which could become
ArchConst.ACheadgroup, in case constants need sharing. That's probably not as
common, you want constants (and types) usually defined in one place, but the
curses case is an exception.