Log state changes at the funnel; never log placeholder statuses

Two small logging fixes from debugging a replicated store. (1) A node could switch consistency mode three ways: an operator broadcast, catch-up at startup, and catch-up triggered by a newer generation. The startup catch-up path changed state silently, so a restarted node's empty data looked like a bug. Put the log line in the one function every path goes through (setMode), log it only when the state actually changes, and write it as 'old -> new' (mode, gen, cleared=true/false). The jump in generation then shows that the node caught up on its own. (2) A request log printed status=ERROR on its 'received' line because the reply struct starts with ERROR as a default. To a reviewer that reads as a failure. Leave status off the received line entirely and write it only when the request is answered. Add a test that rejects 'event=received' lines containing 'status='.