Enable Flink unaligned checkpoints to keep checkpoint duration stable under backpressure

Steps

1. Set execution.checkpointing.unaligned.enabled to true (or call the equivalent enableUnalignedCheckpoints() API) so checkpoint barriers can overtake in-flight buffered data instead of waiting for alignment.
2. Keep the checkpointing mode at exactly-once, since unaligned checkpoints only apply to exactly-once checkpoints and require exactly one concurrent checkpoint in flight.
3. Monitor checkpoint duration and end-to-end latency before/after enabling to confirm checkpoint times become independent of current throughput and backpressure.
4. Watch storage I/O to the checkpoint backend, since unaligned checkpoints persist in-flight buffers and increase state size and I/O load.
5. Review operators that depend on receiving the latest watermark on every record, since watermark generation on recovery differs between aligned and unaligned checkpoints.