Only two buffers ever need to be sychronized and DMA direction of it
is always known based on the type of request we are
processing. Simplify the code by moving synchronization to be a part
of ehci_prepare_qtd(). While at it convert the code to use
dma_map_single() API.

Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>