Continuing the trend of "patches to make Metacity lovers howl", here's
a preference to always put new windows on top, even when they are focus
denied.

The reasoning here is that given sufficiently much real-estate in the
form of large monitors and multihead, the current behavior where if
a window would overlap it is buried and flashed in the taskbar doesn't
work very well - the task bar is no longer in the user's peripheral
vision, and even if they notice it, it can be a long way to mouse to.

(A sophisticated user may know that a single tap of Alt-Tab will get
to the flashing window, but many users will not catch on to this.)

This preference (as noted in the GConf schema) is really meant for
mouse/sloppy focus where we more frequently break the relationship
of focused and on-top. But I don't think it makes sense to make it
a no-op for focus_mode=click. We do break that relationship
occasionally already even in that case, and its less confusing to
try the key and find you don't like it then to have the key do
nothing.

There's a bit of a dependency here on bug 599097 - this patch makes it
even more likely you'll get to the state where a window is under
the pointer but not focused, so having an easy way to recover from that
is nice.

In the case where the new window *completely* obscures the focus
window, it might be nice to focus the no-focus-window to avoid keystrokes
going to a hidden window, but I didn't want to add more code to
complete "completely obscures" just for an obscure preference. It would
be wrong to focus the no-focus-window if the new window is only a small
window that overlaps a bit of the focus window - that basically
reintroduces the annoyance of focus-stealing, if not the security threat.