It's much simpler to just add one extra TRB chained to previous TRB to
handle ZLP. This helps us reduce pointless allocations and simplifies
the code a little bit.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>