[WIP][SPARK-29962][SQL] Avoid changing SMJ to BHJ if one side is non-shuffle and the other side can be broadcast

[wangyum]

What changes were proposed in this pull request?

This PR makes it avoid changing SMJ to BHJ if one side is non-shuffle and the other side can be broadcast because SMJ if faster than BHJ in this case:

spark.range(50000000).selectExpr("id as t11", "id as t12", "id as t13", "id as t14", "id as t15")
  .write.bucketBy(200, "t11").sortBy("t11").saveAsTable("bucketed_table")
spark.range(50000000).selectExpr("id as t21", "id as t22", "id as t23", "id as t24")
  .write.saveAsTable("non_bucketed_table")
spark.sql("DROP TABLE IF EXISTS benchmark_result")
spark.sql("CREATE TABLE benchmark_result USING parquet AS SELECT t1.* FROM bucketed_table t1 JOIN (SELECT * FROM non_bucketed_table WHERE t21 % 5000 = 1) t2 ON t1.t11 = t2.t21")

	SMJ	BHJ
Fisrt time	17.263 seconds	16.556 seconds
Second time	11.948 seconds	13.49 seconds
Third time	14.806 seconds	12.962 seconds

Why are the changes needed?

Adaptive query execution should not degrade performance.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Unit test

[SparkQA]

Test build #114097 has finished for PR 26600 at commit 89ed4e8.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[wangyum]

It is hard to say which one is faster. I'm close it.