This is the latest update of the patch resolving bugs in how device
configurations were reflected in the driver model.  It addresses
the last significant problems I know about in that area.

    - Moves code around so that usb_set_configuration() updates sysfs
      to reflect the current configuration.  Previously, that only
      worked right for the initial configuration chosen by khubd.

        * Previous interfaces are inaccessible.  The code to handle this
          moved from usb_disconnect() into usb_disable_device(), which
          is now called both on disconnect and set_configuration paths.

        * There are new interfaces.  The code to handle this moved
          from usb_new_device() into usb_set_configuration().

        * Resolves a double-refcount problem with USB interfaces,
          by not getting the extra reference in the first place
          and switching to use device_del() to disable interfaces.

        * Comments a similar double-refcount problem with usb
          devices (not interfaces).

      Its kerneldoc is updated appropriately.  The main point being
      that calling usb_set_configuration() in driver probe() methods
      is even more of a no-no, since it'll self-deadlock.

    - Sysfs names for USB interfaces now include the configuation
      number, so that user mode code can't get as easily confused.

      Old style:  "3-1:0" for configs 2 and 3 (interface zero).
      New style:  "3-1:2.0" for config 2, "3-3:3.0" for config 3.

    - Moves usb_new_device() code around a bit, so that the device
      is visible in sysfs before usb_set_configuration() is called.
      (Before the devices for that config's interfaces appear.)

    - Makes the bConfigurationValue be writable through sysfs, so
      device configurations can be easily changed from user mode.
      (Or devices can be de-configured, by setting config 0.)

      There are devices that can benefit from this functionality;
      notably, cdc-acm modems need it right now, so that they can
      be told to use the non-proprietary configuration.  (Since
      the old "change config in probe callback" trick won't work.)