[FEA] Fix Spark 3.1 GpuHashJoin since it now requires CodegenSupport

[tgravescs]

Describe the bug
apache/spark@ae82768
went into Spark 3.1 to add codegen support to ShuffledHashJoin. This commonized some code which added the CodegenSupport trait to the base HashJoin class. We directly inherit from that class in GpuHashJoin so its expecting us to do codegen now.

note that I made the 3.1.0 build work by adding in the missing functions inputRDDs and prepareRelation but this was just to get it to compile again so it wasn't blocking the build of other versions.

Steps/Code to reproduce bug
Please provide a list of steps or a code sample to reproduce the issue.
Avoid posting private or sensitive data.

Run unit tests for Spark 3.1

Expected behavior
A clear and concise description of what you expected to happen.

Tests shouldn't fail or try to run codegen in GPU version.

Environment details (please complete the following information)

• Environment location: [Standalone, YARN, Kubernetes, Cloud(specify cloud provider)]
• Spark configuration settings related to the issue

Additional context
Add any other context about the problem here.