Fast deep-copy-transpose implemented which attempts to not thrash the cache. Good first step for now, but no transpose at all would be preferrable. Started I/O.