drivers: spi_mcux_lpspi: Clean up code

While working on fixing a bug, I found this driver
to be a mess and difficult to read, so clean it up.
Most of the changes have no functional effect, some
however are going to cause minor optimizations in the image.

Trivial changes:
- Organize #includes and #defines and put prototypes at the top.
- Remove some very redundant comments.
- Remove code that was commented out.
- Reword a few comments.
- Add comments to #endif showing the condition ending.
- Remove as many preprocessor directives as possible where
  they are not necessary.
- Reorder some functions to be located near other relevant
  functions (eg, put init and isr near the device init macro).
- Add some delimiting comments for readability.
- Remove many very unnecessary newlines.
- Align slashes at the end of macros.
- Align debugging statements.

Minor Changes:
- Consolidate some duplicated code in the dma paths into
  helper functions (spi_mcux_dma_equal_block_length and
  spi_mcux_dma_common_load). This should make the image
  smaller but have no functional difference besides a struct
  potentially getting it's members filled in a different order.
- Reduce redundancy and ifdef chaos in the transceive wrappers
  by making another wrapper function.
- Some source code changed to reduce indentation levels
  and nested conditional controls which were hard to follow,
  but functional logic is kept the same.
- Remove absolutely redundant and pointless wrapper of k_spin_lock.
- Remove parent_dev from config struct and utilize irq_config_func
  in all cases, with a different definition for lp_flexcomm / legacy.

Fixes:
- Fix some code added in 17032a0
  that appears to have a bug where transceive_dma is called even
  if there is no dma device from spi_mcux_transceive_async.
- Fix a code path in transceive_dma that appears to have a control
  flow bug (probably due to the super hard to follow nesting and
  #ifdefs), where under a certain scenario if an error happened,
  some code that seems to depend on the earlier code would execute,
  but the earlier code didn't execute because of the error,
  seems like what was meant to happen was to wrap up and exit the function
  in case of error like the other code path.
- Fix the slave select input check allowing slave 4, because of using
  > instead of >= in the check.

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