Faster, more reliable Kura replication mesh recovery and node bootstrap

Kura replication is now more resilient during a full-mesh cold-start. A fresh or rebuilt regional node can bootstrap the whole cache in minutes instead of hours, or failing to finish at all, because artifact bodies are fetched in parallel rather than one at a time. A serial fetch left the network idle between transfers and could not complete a large cache within the bootstrap timeout. The peer-to-peer transfer timeout also changed from a fixed total request timeout to a read timeout, so large artifacts that transfer slowly but steadily no longer get cut off mid-stream and wedge the mesh. As a hardening guard, the serving path also verifies a segment file is actually long enough for the requested range before streaming it, so a truncated file returns an error rather than silently serving a短