[GLUTEN-3654][VL]Fix the duplicated key exception in TopN

[JkSelf]

What changes were proposed in this pull request?

The TakeOrderedAndProjectExec will be transformed into either a sort + limit or directly into a limit operation. If it is a sort + limit case, Velox will convert it into a TopNNode. Here, we need to do some validation on the TopNNode.

How was this patch tested?

Pass Jenkins test

[github-actions]

#3654

[github-actions]

Run Gluten Clickhouse CI

[github-actions]

Run Gluten Clickhouse CI

[JkSelf]

@rui-mo Passed Jenkins TPC-DS test. http://sr168:8080/view/Gluten/job/Gluten_TPCDS_test/874/

[zhouyuan]

@JkSelf it looks like the tpcds test in GHA is fine, is this due to the difference on scale factor?

-yuan

[JkSelf]

@JkSelf it looks like the tpcds test in GHA is fine, is this due to the difference on scale factor?

-yuan

@zhouyuan It seems that this issue is associated with the type of Spark application submission. I can reproduce this problem using spark-shell with a data scale of 1GB.

[GlutenPerfBot]

===== Performance report for TPCH SF2000 with Velox backend, for reference only ====

query	log/native_3655_time.csv	log/native_master_11_09_2023_745ecb383_time.csv	difference	percentage
q1	33.93	33.80	-0.128	99.62%
q2	24.71	24.95	0.236	100.95%
q3	40.03	38.87	-1.162	97.10%
q4	35.38	36.96	1.579	104.46%
q5	69.43	69.94	0.514	100.74%
q6	7.91	7.87	-0.036	99.54%
q7	84.98	85.40	0.421	100.49%
q8	88.07	88.61	0.543	100.62%
q9	122.43	121.14	-1.294	98.94%
q10	47.55	54.35	6.800	114.30%
q11	19.99	19.52	-0.476	97.62%
q12	24.81	24.93	0.121	100.49%
q13	48.91	50.92	2.010	104.11%
q14	16.78	18.17	1.387	108.27%
q15	30.26	30.79	0.534	101.77%
q16	15.97	16.06	0.093	100.59%
q17	104.07	101.51	-2.557	97.54%
q18	149.32	148.23	-1.086	99.27%
q19	14.85	14.95	0.102	100.69%
q20	30.61	31.24	0.623	102.03%
q21	223.74	223.30	-0.440	99.80%
q22	13.57	14.24	0.670	104.94%
total	1247.28	1255.74	8.454	100.68%