We currently don't advance the co_positions iterator for CACHE entries, which can lead to incorrect positions being reported for later instructions. Also, the positions for cache entries should be populated, not None.

Also, don't show a "pretty" argrepr for cache entries unless they're actually being used by a quickened instruction. This cuts down on visual clutter when show_caches=True, adaptive=False, or for instructionss we've completely unquickened.