drivers: spi: spi_sam0: fix spi_sam0_fast_txrx()

The optimisation in `spi_sam0_fast_txrx()` is broken, loading
two bytes into the `DATA` register in rapid succession will lose
one byte.

This can be observed by running `tests/drivers/spi/spi_loopback`.
The test will get stuck in `spi_sam0_fast_txrx()` forever waiting
for the final byte.

Undo this small optimisation and only load the next byte into the
`DATA` register after the response has been received.

This fixes `tests/drivers/spi/spi_loopback`.

Signed-off-by: Benjamin Valentin <benpicco@googlemail.com>