<machine/endian.h>: Use __builtin_bswapXY function if c++.

<machine/endian.h>: Use __builtin_bswapXY function if c++.

 There builtins seem to be supported since gcc 4.8.  In theory we could
 use them for C code too, but for now limit this to c++ so to help
 compilers to deal with fun what is constexpr, lambdas, templates, etc.
 Generally, this reduces amount of debug code created through the
 __word_swap_foo() expansions too.

 Note: g++8 implements __word_swap_foo_var() with a bit different asm.