Fix Left join implementation is incorrect for 0 or multiple batches on the right side

[Dandandan]

Which issue does this PR close?

Closes #235
Closes #239
Closes #143

Rationale for this change

Fixes behavior of left join with regard to multiple batches on the right side and 0 right side batches.

This way it also also likely much faster as it avoids keeping generating / keeping / indexing into a HashSet for each right side batch.

What changes are included in this PR?

• We add a Vec<bool> to keep track of left-side rows that didn't match with the right side. (number o left rows bytes extra memory usage for left joins, not that much compared to storing the left side data and hashmap + indices. It could be more memory-efficient by using a bitmap instead.
• Generate rows for any item that is not marked as visited.
• Add some tests for 0 / multiple batches on the right

Are there any user-facing changes?

Only correctness improvements

[codecov-commenter]

Codecov Report

Merging #238 (ab760af) into master (c945b03) will increase coverage by 0.30%.
The diff coverage is 96.42%.

@@            Coverage Diff             @@
##           master     #238      +/-   ##
==========================================
+ Coverage   76.46%   76.76%   +0.30%     
==========================================
  Files         135      134       -1     
  Lines       23250    23248       -2     
==========================================
+ Hits        17777    17847      +70     
+ Misses       5473     5401      -72     

Impacted Files	Coverage Δ
datafusion/src/physical_plan/hash_utils.rs	100.00% <ø> (ø)
datafusion/src/physical_plan/hash_join.rs	87.21% <96.42%> (+1.27%)	⬆️
ballista/rust/core/src/client.rs	0.00% <0.00%> (ø)
ballista/rust/executor/src/main.rs	0.00% <0.00%> (ø)
ballista/rust/scheduler/src/main.rs	0.00% <0.00%> (ø)
ballista/rust/executor/src/execution_loop.rs	0.00% <0.00%> (ø)
ballista/rust/executor/src/flight_service.rs	0.00% <0.00%> (ø)
...ust/core/src/execution_plans/unresolved_shuffle.rs	50.00% <0.00%> (ø)
ballista/rust/executor/src/lib.rs
ballista/rust/scheduler/src/lib.rs	21.13% <0.00%> (+1.62%)	⬆️
... and 1 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c945b03...ab760af. Read the comment docs.

[andygrove]

LGTM. I pulled this branch locally and confirmed that this fixes the failing tests. Thanks, @Dandandan!

[jorgecarleitao]

Yap, the implementation makes a lot of sense. Thanks a lot @Dandandan !

[Dandandan]

Merging this when it's green