DIRECTORY calling DELETE-DUPLICATES is slow

I waited about 20 minutes for DIRECTORY to finish getting the
pathnames of a large directory containing about 27000 files.  The
problem is that DIRECTORY calls DELETE-DUPLICATES which is implemented
using an order n-squared algorithm!  For large sequences, the use of a
hash-table would make sense, but would cause consing.

I better change would be to the DIRECTORY function itself.  Note that
the result of DELETE-DUPLICATES is passed to SORT.  If the list were
sorted first, then it is easy (order n) to remove duplicates.