This includes:

 * Adding a test for bswap16 with a low byte >= 128
 * Also test the byteSwapN functions from GHC.Word, tested both INLINE
   and not INLINE, so we test both independent parts: the compilation of
   base, and the backend compiling the code *using* base.
 * Fix the usage of byteSwapN# primitives in the test, by masking off
 * the higher bits when storing the results in Word16/Word32.

Thanks to Reid Barton for the investigation.

Authored-by: Reid Barton <rwbarton@gmail.com>
Signed-off-by: Austin Seipp <austin@well-typed.com>