Patch for PR other/13906 (md5.c does unaligned memory accesses)

[DJ Delorie]

> Well, if the alignment of the md5_uint32 type was greater than the size
> of the md5_uint32 type, then buffer wouldn't necessarily have to have
> the same alignment.  I doubt there are any real platforms with this
> situation.

Let's hope not, but OK, check it in.