[FEA] Implement `outputPartitioning` for GPU join execs

[andygrove]

Is your feature request related to a problem? Please describe.

Spark's BroadcastHashJoinExec, ShuffleHashJoinExec, and BroadcastNestedLoopJoinExec classes implement outputPartitioning, but our GPU implementations do not. This could potentially lead to missed optimizations.

Describe the solution you'd like

• Add failing tests to compare GPU vs CPU join plans to ensure they have the same output partitioning
• Implement outputPartitioning

Describe alternatives you've considered

Additional context