[Bug-gnubg] another minor analysis bug

[O.M. Riordan]

If you analyse on some depth, select the best move in some position,
re-analyse it on lower depth, then on the original depth, it does not go
back to the top of the list. This then messes up the statistics, as the
error for this move is measured w.r.t. the second best move, and so may
well be a large negative number. 

Oliver