kernel - u4b - Implement doorbell

kernel - u4b - Implement doorbell

* Implement the doorbell properly.

* Use the doorbell to clear the previous transfer when reusing a transfer.
  This is a horribly inefficient hack but it's better than blowing up the
  controller

  (This is in addition to the xfer freeing fifo hack)

* We also seem to need to ring the doorbell when queueing transfers.  This
  should not be necessary.

* Enable ASP.  This allows qTD bursts, making high speed busses more
  efficient.

* NOTE:  There are still problems with umass.  With some USB sticks,
 	 writing to the stick causes the cpu to stall for very long
	 periods of time.  It is shown as 'interrupt' overhead but insofar
	 as I can tell it's actually the controller stalling the memory
	 bus for reasons unknown.

	 This does not happen with e.g. a SATA SSD connected through a USB
	 adapter.