Using Semantic Kernel with max_round_count set to limit iterations stops execution correctly, but instead of returning the partial results accumulated so far, the output is simply: "max round reach" ...