Restart and replication with maintainers (V18) #330

Part of epic #305.

Problem. Startup loads every KG sequentially and rebuilds nothing derived (src/storage_engine/mod.rs:1754-1916); the follower re-evaluates standing queries on its own snapshots (PR #283 src/replication/follower.rs:424-462).

Target. At load, start maintainers and build generations in parallel across KGs (bounded by CPUs); a KG is queryable when its frontier reaches the loaded revision; the follower feeds its maintainer from applied events (no derived state shipped); /v1/replication/status reports the follower's frontier by primary revision; optional later: serialized trace batches as arrangement checkpoints next to shards (separate issue when needed).

Acceptance. B14 (views ready <= 2x today's load time at 1M edges); kill-9 and restart keeps every subscription's resubscribe snapshot equal to recompute; follower views equal primary views at the same primary revision in the EN-12 harness.

Size M. Dependencies V2, #135 for parking. Order slot 6.