Fix Kura replication mesh deadlocks and speed up fresh-node bootstrap

The Kura replication mesh no longer deadlocks during a full-mesh cold-start. Previously, a 30-second total request timeout could abort large artifact transfers mid-stream, leaving peers stuck with incomplete bodies and preventing any node from becoming ready. The peer client now uses an idle read timeout instead of a total timeout, so slow but progressing transfers complete while genuinely stalled connections still fail fast. Bootstrap also fetches artifact bodies concurrently instead of serially, so a fresh or rebuilt regional node can replicate the full dataset in minutes rather than hours or never.