Ensure the hash_file and hash_dir routines give identical results
on big- and little-endian systems. The default hash routines for
integer types are endian-dependent, so all such hash inputs need
to be byte-swapped.
This applies in particular to the file hashes used as input when
computing directory hashes.
In addition, the default hash routines for composite types use
a length prefix, which it itself an integer type (usize). In
order to be able to byte-swap that prefix, we have to re-implement
those bits of the standard library ourselves.