Treat the populate reply wait as an idle timeout so large indexes can finish

[mussbach]

In QueuePagerPersister::insert() the page-messages are all dispatched first, then the sender waits for replies with limit_overall_reply_time (default 180s) enforced as a hard overall cap on the reply-draining phase.

That cap is fine when the consumers clear the backlog quickly, but it trips for large indexes: if the consumers need more than 180s to drain after dispatch (slow/contended cluster, modest consumer count, big index), insert() aborts with Overall reply time … exceeded even though the consumers are healthy and still replying — leaving the populate failed but the index partially written.

This reworks the cap into an idle timeout: the deadline is pushed forward on every received reply, so a steady stream of replies completes regardless of total runtime, while a genuine stall (all consumers dead → no reply within the window) still aborts. The exception message is updated to reflect the new meaning.

Single-file change to the reply loop; no API or config change. limit_overall_reply_time now behaves as a per-reply idle window rather than a wall-clock budget for the whole drain.