Java bindings for cudf::hash_join

[jlowe]

Adds Java APIs to build a hash table for the right side table in a join and re-use it to join against a series of left probe tables. It also exposes the ability to compute the join output row count and pass that count to a subsequent call to produce the join gather maps to avoid redundant computation when the output row count must be examined before manifesting the gather maps.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.10@8b02ca3). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-21.10    #9080   +/-   ##
===============================================
  Coverage                ?   10.71%           
===============================================
  Files                   ?      114           
  Lines                   ?    19093           
  Branches                ?        0           
===============================================
  Hits                    ?     2046           
  Misses                  ?    17047           
  Partials                ?        0           

---

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 8b02ca3...f3480fd. Read the comment docs.

[revans2]

The code looks fine. I would love to see if we could some how make this spillable, and how expensive spilling would be compared to re-building it. For broadcast joins the size of the build table should be small enough that it would be ideal to keep the table in memory all the time, but that would require that the build side Spark picks matches the build side that cudf supports.

[jlowe]

@gpucibot merge