[multistage] Bushy Parallelism for Join Node Inputs

At present we first the process the right sub-tree of a join to build a HashTable and only after that's done do we process the left sub-tree.

Ideally we should be able to process the two sub-trees concurrently (bushy parallelism).

cc: @walterddr

Start by tracing the join node execution path for the right and left sub-trees, focusing on where the right side builds the HashTable before the left side is processed. Done means both sub-trees can be processed concurrently while preserving correct join results; the issue names no files or tests to run.

Written by the indexing model from the issue text.