There is a little bug with parallel scanning: a two-shards partition is a "single partitioned stage", so KqpTaskGraph generates 1 task for ReadActor, one RA reads two shards, but RA can only return one row as BatchOperationMaxRow and it is not clear how to determine which row to return, from the first shard or from the second shard?
Solution: let's make a sequential scanning by selection only the row from the first shard.