Kura bootstrap no longer amplifies disk and network when peers share artifacts

Kura nodes now bootstrap without amplifying disk or network usage when multiple peers serve the same artifacts. Previously, a fresh node fetching from N peers with shared keys could write ~Nx the dataset to disk and exhaust its volume (ENOSPC on a 50Gi PVC in staging), and also downloaded the bodies once per peer (~4x WAN transfer for a 9.2G dataset across 4 peers). After this fix, concurrent applies for the same artifact id are deduped so only one copy is written, and per-artifact bootstrap fetch gates single-flight the body download across peers. Operators should see new nodes reach Ready reliably on standard disk sizing, with a single copy of shared artifacts transferred and stored instead of one per peer.