A collection of mostly orthogonal and small fixes:

 - Update various corner cases in the side suspend/resume code,
   triggered mostly on the host side:
     * avoid spurious gadget driver resume callbacks in B_IDLE state
     * avoid spurious gadget driver suspend callbacks in A_HOST state
     * cope better with some spurious resume irqs
     * partial fix for disconnection issues in A_SUSPEND

 - Resolve some TUSB rmmod races by disabling IRQs at the proper moment,
   and killing the idle timer.

 - Only mark TUSB device as wakeup-capable if it's connected to an
   IRQ which allows that.

 - Don't idle the TUSB chip until khubd handles all pending port
   status change events; dump that status in procfs.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>