one batch of changes completed

[Havoc Pennington <hp redhat com>]

Hi,

HEAD should be relatively stable again. The tests pass anyway.

I changed the IDL to use a ConfigDatabase object instead of the
ConfigServer_Context integer ID, and split gconfd.c into two files
(added gconf-database.c), which makes it a bit more manageable.

This allowed a reasonably good implementation of restoring registered
listeners from the log file. I have seen gconfd exit, come back up,
and listeners still work for apps that were running prior to the
gconfd exit, anyway. This is just a first cut; there are robustness
issues that could be improved.

Another advantage of the ConfigDatabase object is that gconfd could
actually provide a ConfigDatabase object running in a non-gconfd
process. So it could hand off work directly to a backend-specific
process, for example, or distribute load among various daemons. I'm
not implementing this anytime soon, but it's a nice conceptual
feature.

Most likely I've introduced a few "doh!" bugs, so if you see them, 
let me know...

Havoc