• [Enhancement] Report detailed blocking information (active listeners, alive workers, and in-processing jobs) during forceful shutdown instead of only aggregate counts.
• [Enhancement] Improve ForcefulShutdownError description to clearly explain when and why it is raised.
• [Enhancement] Cache messages.last in BatchMetadata builder to avoid duplicate array traversal.
• [Enhancement] Optimize VirtualOffsetManager#mark to use a single array scan instead of separate include? and index calls (Pro).
• [Enhancement] Optimize VirtualOffsetManager#materialize_real_offset to use keys.sort instead of to_a.sort_by with tuple destructuring (Pro).
• [Enhancement] Optimize IntervalRunner#call to use a single monotonic_now call instead of two per invocation.
• [Enhancement] Support WaterDrop :fd mode in Swarm.
• [Maintenance] Use both :fd and :thread producer backends in CI.
• [Maintenance] Include spec file hash in integration test topic names for easier traceability in Kafka logs (#3056).
• [Fix] Remove duplicate topic creation in multi-broker health integration specs (#3056).
• [Fix] Preserve producer-specific kafka settings (e.g., enable.idempotence) when recreating the producer in swarm forks.