Updates the PCI glue for HCDs:

 - There's now just one code path to suspend. CONFIG_USB_SUSPEND mostly
   means bypassing the "idle the root hub" part, and it uses the same
   D1/D2 logic when necessary.

 - Though usbcore is moving away from device->power.power_state, it still
   needs to be initialized correctly (else pmcore can get confused).

 - Disable the device on init path failures.

The first two address real bugs, the third was just something I noticed
while fixing the second.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>