Fix address generation for writes + generalize to arbitrary-dimension arrays

[ethanuppal]

This issue is notes for me to execute next. The current state of #24 has address generation unequivocally shift back the program, which is invalid for writing memory (as the address must be available at the exact same time as the write happens). To fix this, we can provide a general algorithm that

• Creates a new thread for each access location
• Performs a time travel, cascading up the tree
• Schedules these threads together in the address generator

(You've already worked this out; see Slack)

(Of course, this needs to be made more efficient in area)