spi_nxp_lpspi: Fix fifo fill logic

Fixing some issues with the TX fifo fill logic in two places:

First in the normal fill function, it didn't take into account a
situation where the TX fifo is already partially filled. This currently
doesn't cause a problem because the driver is written in a way that the
watermark is always 0 for TDF, but in case the watermark were anything
else it would cause a problem.

Second, when filling the TX fifo with NOPS in order to clock the rest of
the RX in from the bus, the calculation regarding the current TX fifo
length was just wrong and was leading to a bug in some cases where there
was a subtraction underflow and billions of NOPs were being filled.
Also, there could be a problem where a few extra NOPs are put in the TX
fifo if we don't count what we already have in the TX fifo, so fixing
that as well.

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