fix(shared-log): harden rebalance repair under churn

[marcus-pousette]

Summary

• force a fresh rebalance pass on removal-driven churn by allowing toRebalance(..., { forceFresh: true })
• proactively evict remote replicator ranges on unsubscribe to avoid stale-leader selection when unsubscribe races replication reset messages
• add a full repair sweep + bounded retry scheduling after removal churn
• chunk force-fresh repair dispatches so churn recovery uses deterministic simple-sync batches instead of a single large IBLT pass

Tests

• pnpm -C packages/programs/data/shared-log test -- --grep "maturity will retrigger rebalance|handles peer joining and leaving multiple times" --reporter=min
• stress loop: 8x pnpm -C packages/programs/data/shared-log test -- --grep "handles peer joining and leaving multiple times" --reporter=min
• pnpm -C packages/programs/data/shared-log test -- --grep "u64-iblt" --reporter=min