[AVR32] Implement dma_{alloc,free}_writecombine()

Implement dma_alloc_writecombine() and its dma_free_writecombine()
counterpart. These will do basically the same thing as
dma_alloc_coherent() except that the virtual mapping will allow
write buffering, causing better performance for certain use cases
like frame buffers.

The same API is already available on ARM.

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>