index.c, instead of storing a distinct tree root for every entry in

the trie, now stores a distinct tree root for only those entries
we're actually going to want to look up later - i.e. those at the
start or end of a directory interval. This means that nodes can be
modified and reused by insertions between two such points, which
means we don't have nearly so many duplicate nodes and save a lot of
space in the index without losing any functionality. This leads to a
_huge_ improvement on the size of the index file: on my home
directory it gets a factor-of-5 improvement from 300Mb to 60Mb!

git-svn-id: svn://svn.tartarus.org/sgt/agedu@8286 cda61777-01e9-0310-a592-d414129be87e