Implement Flink event-time windowing with watermarks and handle late records via side outputs

Steps

1. Assign timestamps and watermarks using stream.assignTimestampsAndWatermarks(WatermarkStrategy.forBoundedOutOfOrderness(Duration.ofSeconds(10)).withTimestampAssigner(...))
2. Apply a time window (TumblingEventTimeWindows, SlidingEventTimeWindows, or EventTimeSessionWindows) after keyBy() using .window(TumblingEventTimeWindows.of(Time.minutes(5)))
3. Implement a ProcessWindowFunction or aggregate using reduce()/aggregate() to compute the window result when the watermark passes the window's end timestamp
4. Define a side output tag: OutputTag<T> lateOutputTag = new OutputTag<T>('late-data') {}; and attach it to the windowed stream with .sideOutputLateData(lateOutputTag)
5. Retrieve the late record stream using windowedStream.getSideOutput(lateOutputTag) and route late records to a dead-letter sink or correction pipeline
6. Set allowedLateness(Time.minutes(1)) on the window to keep window state open after the watermark passes, emitting updated results for late-but-not-too-late records