spi_nxp_lpspi: Fix DMA driver async return

We should not release context until transfer ends. The code previously
would return from wait_for_completion and then release the context. This
is only supposed to be done in the dma callback except for the case of
error in the transceive call. For async transfer this was most likely
always happening wrong and probably broken for multi threads trying to
access the bus due to this premature release of the context.

Also we should not enable CS and leave enabled in case of error, move CS
enable to after the error check.

Signed-off-by: Declan Snyder <declan.snyder@nxp.com>