Implement address generator synthesis

[ethanuppal]

Implements #25.

[ethanuppal]

Currently this is very naive and incomplete. Here's a nonexhaustive list of things to do:

• Have TimingAnalysis do a kind of "polyhedral analysis" and determine latency of loops with non constant integer bounds.
• Instead of inserting delay at the start, see if you can do something smarter that's more susceptible to par-CFG merges.
• And of course, clean up the code.

[ethanuppal]

Would it be possible to just separate out the parts of the program that compute the addresses explicitly?