Refactor interpreted code path + fix nulls in join keys

[Tom-Newton]

• Added some more tests around nulls in the join keys and duplicate input rows.
  • Fixed both codegen and interpreted code paths for nulls in join keys.
• Refactor interpreted code path.
  • Remove lots of legacy from the normal spark sort merge join related to buffering multiple matches for each streamed row. With the PIT join we only allow one match per left row (this does create some non-determinism if there are duplicate input rows).
• Update comments and renamed lots of variables
  • Rename streamed -> left and buffered -> right. Streamed and buffered only really made sense in the normal sort merge join because of the buffering multiple matches thing. Additionally in the PIT join the left and right dataframes are considered fundamentally different.