feat: Investigate potential issues with HashJoin and LeftSemi #2697

What is the problem the feature request solves?

When having

      "spark.comet.exec.replaceSortMergeJoin" -> "true",
      "spark.sql.adaptive.enabled" -> "false",

The TPCDS Q69 giving correctness issues like analyzed in #2667
Currently we stop join rewrite but it can be a bigger problem behind and we need to investigate the rootcause

Describe the potential solution

No response

Additional context

No response