More loop-body simplifications.

Rationale:
This removes all dead induction from the CaffeineLogic loop,
giving yet the next performance boost  (2700us->1700us).
Also, the runtime is now the same between a DX compiled
and JACK compiled version, giving confidence that all
recent introduced optimizations are generally useful
and something expected from any optimizing compiler.
Last, less realistic improvement will pale anything
seen so far, since it removes the full loop (still TBD).

Test: test-art-host
Change-Id: Id6b89f74b7d009616821dca195200933cc0eaaf2