While investigating #142 (closed) an optimization opportunity came to mind. The number of codes-mapping calls in the dragonfly can be reduced by a significant amount by storing the router LP-ID in the terminal rather than the relative ID, which then must be re-calculated for each flit.

We could take this one step further by storing the full set of neighbor LP IDs, since the neighbors of any dragonfly entity are statically defined (could do this for any statically-defined network topology), but those changes would be more far-reaching.