Cleanups:
o  Log messages should show both original and new size of the data. The speed
   should be probably calculated from the former.
o  Buffer sizing in shep-export.

Improvements:
o  When quicksorting a large input (especially in threaded case), invest more
   time to picking a good pivot.
o  Overlay presorter I/O with internal sorting.

Users of ucw/sorter/array.h which might use radix-sorting:
indexer/chewer.c
indexer/lexfreq.c
indexer/mkgraph.c
indexer/reftexts.c