A save waiting on a shutdown no longer waits for nobody

When a node is shutting down while one of your saves is still in flight, the answer is deliberately held back and sent from a later step in the shutdown instead. There is a good reason: that later route still reaches your session, and an answer sent from the earlier one might not.

But nothing checked whether your session was still listening. If it was not — because it had already been answered, because the wait had run out, or because the save came from a different process entirely — the answer was held back for a route that would deliver it to nobody, and your save was left with silence until its own window ran out. The hold-back now happens only when there is somewhere for the answer to go. Otherwise the node answers immediately, on whatever route is still open.

There was a second, quieter problem behind it. An answer that arrives after the window closes is not delivered — past that point it cannot be told apart from a stale repeat of an older one, and acting on it would be worse than ignoring it. That part is unchanged and deliberate. What has changed is that it used to be completely silent, and indistinguishable from an answer that was never produced at all. Those are two different problems with two different causes, and the logs showed the same nothing for both. A late answer is now recorded, with how late it was.

Nothing waits longer as a result — no window was widened. A save that could be answered is answered sooner, and when one genuinely cannot be, the reason is now written down instead of inferred.