Reduce the N in the logN term for the sorting cost associated with computing

Reduce the N in the logN term for the sorting cost associated with computing
DISTINCT by B-Tree by half, under the assumption that the DISTINCT will
eliminate about half the rows of output.  This is really a wild guess.  But
we do not have any better way of estimating what the row-count reduction due
to DISTINCT will actually be.

FossilOrigin-Name: 8787417ec1da8071d84c6ff0d7a90b5fd458ab6baba871327f36bc4e1bceca61