currently multi-way merges during index construction
and update are handled by scanning all entries. A heap
is the classic computer science solution for this. It
may not actually be better due to overhead imposed by
the generic heap structure, so test+time to make sure.
This should allow deferral of merging for greater
numbers of temporary files, which should be a speedup.
It will consume more disk space, test to see how much
and decide what is a good level.