reduce pipelined DML RPC count in some case to optimize performance

[crazycs520]

Enhancement

#50389 introduce pipelined DML, we can enhancement the performance in some situation. See following test:

func TestPipelinedDMLInsertRPC(t *testing.T) {
	store := realtikvtest.CreateMockStoreAndSetup(t)
	tk := testkit.NewTestKit(t, store)
	tk.MustExec("use test")
	tk.MustExec("drop table if exists t1")
	tk.MustExec("create table t1 (a int, b int, unique index idx(b))")
	res := tk.MustQuery("explain analyze insert ignore into t1 values (1,1), (2,2), (3,3), (4,4), (5,5)")
	explain := getExplainResult(res)
	require.Regexp(t, "Insert.* check_insert: {total_time: .* rpc:{BatchGet:{num_rpc:1, total_time:.*}}}.*", explain)
	tk.MustExec("set session tidb_dml_type = bulk")
	res = tk.MustQuery("explain analyze insert ignore into t1 values (1,1), (2,2), (3,3), (4,4), (5,5)")
	explain = getExplainResult(res)
	// TODO: try to optimize the rpc count, when use bulk dml, insert ignore will use 5 BufferBatchGet and  1 BatchGet rpcs.
	// but without bulk dml, it will only use 1 BatchGet rpcs.
	require.Regexp(t, "Insert.* check_insert: {total_time: .* rpc:{.*BufferBatchGet:{num_rpc:5, total_time:.*}}}.*", explain)
	require.Regexp(t, "Insert.* check_insert: {total_time: .* rpc:{.*BatchGet:{num_rpc:1, total_time:.*}}}.*", explain)
	tk.MustExec("delete from t1")
}

func getExplainResult(res *testkit.Result) string {
	resBuff := bytes.NewBufferString("")
	for _, row := range res.Rows() {
		_, _ = fmt.Fprintf(resBuff, "%s\t", row)
	}
	return resBuff.String()
}