Release v19.29.2

Changed

• SQL read-model storage records completed replay promotions in a chronicle_replay_promotions table in each read-model database. A replay that would overwrite an unrecognized backup table now stops with an error instead of dropping it; on PostgreSQL, an older backup whose shortened name collides with a new one is kept as chronicle_replay_backup_<id> and logged for manual cleanup (#4476)

Fixed

• Finishing a replay of a SQL read model no longer fails, or briefly exposes a missing table, when the read model is queried or written while the replayed table is promoted; the swap now commits atomically and is safe to retry on PostgreSQL and SQL Server (#4476)