[NSE-45] BHJ memory leak

[zhztheplayer]

#45

This is ideally to fix leaks and reduce process RSS during BHJ-related query execution. Still doing a final check for memory segmentation and other minor stuffs.

Checklist:

1. BHJ: Fix unclosed hash map on driver
2. BHJ: Fix unclosed input batches on driver
3. Remove use of unreliable JVM API finalize()
4. Remove ref-count for java-side hash maps (we should avoid creating more manual ref-count logic in JVM code unless performance getting really worse)
5. Remove option broadcastCacheTimeout which is no longer needed.

[xuechendi]

@zhztheplayer , codes are looking good to me, let me verify with a tpcds run.

[xuechendi]

@zhouyuan, seems this PR will fail in my setup tpcds sf500 dpp test, just synced with @zhztheplayer , he will continue to verify with TPCDS

[zhztheplayer]

Thanks @xuechendi Let's hold off until everything gets stable.

[github-actions]

[github-actions]

[github-actions]