Parameterize systolic generator to use different PE

[rachitnigam]

Add an option to the systolic array generator to use a specific mac_pe implementation. This will allow us to, say, replace the default mac_pe implementation which uses sequential multipliers with unsynthesizable ones so that they can be simulated with tools like the symbolic evaluator.

[rachitnigam]

@calebmkim we briefly talked about implementing a weight-stationary systolic array. If we end up doing that for the camera-ready version of the paper, let's close this issue!