We used to miss a lot of conflicts during counting in `logBackjump`.
Counting initial conflicts here drastically reduces search time in some
cases.
Not sure whether counting the current conflicts (`cs`) may be even
better than counting the initial conflicts (`initial`).