Faster Kura cold-bootstrap write throughput via group-committed segment fsyncs

Fixes a write-path serialization bottleneck in the Kura store: every artifact write held the global segment lock across an fsync, so concurrent artifact saves and inbound peer replication serialized at one fsync per write, capping write throughput at roughly 1 / fsync_latency. Writes now group-commit their durability fsync, so N concurrent writes (for example the parallel bootstrap fetches a fresh node drives) share ~1 fsync instead of N. Result: a fresh Kura node finishes cold-bootstrap materially faster, and a degraded mesh re-converges faster, with the win growing under more concurrent inbound writes.