[GLUTEN-4713][CORE] Fix invalid children caused by std::move in RowVectorStream

[liujiayi771]

What changes were proposed in this pull request?

In #4713, we discovered that executing q23b on large scale TPCDS dataset would result in a core dump. After investigation, we concluded that an optimization introduced in #4628, which removed some unnecessary post-projects, led to the occurrence of the core dump. However, we believe that this is only a symptom and not the root cause of the problem. In #4726, we restored the post-project and continued to investigate the reason behind the core dump that was triggered by the absence of the post-project.

We ultimately identified the root cause of the issue to be that the next method of RowVectorStream moves away the children of RowVector, leaving the RowVector's children in an indeterminate state. Since the aggregation operation reuses the output RowVector, accessing a child that has been moved away during reuse can lead to a core dump.

1. Why doesn't a core dump occur when there is a post-project?
   FilterProject operator in Velox does not reuse the output RowVector, it always creates a new RowVector to return.

2. Why doesn't the bug get triggered on small datasets?
   The bug is only triggered when there is a reuse scenario occurring with multiple batches being outputted by the aggregation. With small data, the aggregation would only output a single batch, hence no issue arises. However, if the batch size is reduced, this bug can also be triggered.

3. Does the bug occur when the aggregation outputs multiple batches?
   The bug is triggered when the downstream of the aggregation is a RowVectorStream operator.

After the fix, we can revert the optimization of removing unnecessary post-projects introduced in #4628.

How was this patch tested?

Testing 1T TPCDS q23b.

[github-actions]

#4713

[github-actions]

Run Gluten Clickhouse CI

[liujiayi771]

cc @zhouyuan, @rui-mo, @ulysses-you, thanks.

[rui-mo]

Thanks! Verified on internal TPC-DS test.

[ulysses-you]

lgtm, thank you

[ulysses-you]

/Benchmark Velox

[GlutenPerfBot]

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

query	log/native_4753_time.csv	log/native_master_02_23_2024_9ceade94a_time.csv	difference	percentage
q1	33.78	34.69	0.909	102.69%
q2	22.67	24.33	1.668	107.36%
q3	37.82	38.86	1.041	102.75%
q4	37.78	36.08	-1.697	95.51%
q5	70.88	72.24	1.362	101.92%
q6	7.42	7.31	-0.110	98.52%
q7	83.65	84.62	0.973	101.16%
q8	85.91	85.39	-0.511	99.41%
q9	121.99	124.53	2.545	102.09%
q10	45.19	44.23	-0.963	97.87%
q11	20.41	19.94	-0.468	97.71%
q12	28.69	26.67	-2.017	92.97%
q13	45.48	44.33	-1.160	97.45%
q14	17.98	20.85	2.868	115.95%
q15	27.56	29.04	1.475	105.35%
q16	14.30	14.19	-0.115	99.20%
q17	102.03	102.20	0.172	100.17%
q18	149.50	150.35	0.850	100.57%
q19	15.50	14.06	-1.441	90.70%
q20	27.07	26.19	-0.879	96.75%
q21	224.87	226.58	1.706	100.76%
q22	13.86	13.90	0.044	100.32%
total	1234.32	1240.58	6.254	100.51%

[GlutenPerfBot]

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

query	log/native_4753_time.csv	log/native_master_02_23_2024_9ceade94a_time.csv	difference	percentage
q1	32.45	34.69	2.240	106.90%
q2	25.46	24.33	-1.131	95.56%
q3	37.69	38.86	1.171	103.11%
q4	38.84	36.08	-2.760	92.89%
q5	70.38	72.24	1.867	102.65%
q6	7.04	7.31	0.267	103.79%
q7	87.27	84.62	-2.656	96.96%
q8	85.60	85.39	-0.202	99.76%
q9	122.22	124.53	2.306	101.89%
q10	43.20	44.23	1.022	102.37%
q11	20.14	19.94	-0.204	98.99%
q12	26.50	26.67	0.169	100.64%
q13	45.40	44.33	-1.074	97.63%
q14	17.85	20.85	3.004	116.83%
q15	29.93	29.04	-0.893	97.02%
q16	13.91	14.19	0.273	101.97%
q17	102.99	102.20	-0.789	99.23%
q18	150.78	150.35	-0.425	99.72%
q19	13.77	14.06	0.288	102.09%
q20	28.13	26.19	-1.941	93.10%
q21	227.60	226.58	-1.025	99.55%
q22	13.66	13.90	0.235	101.72%
total	1240.83	1240.58	-0.259	99.98%