Auron Flink native operator support watermark

[Tartarus0zm]

We've added support for Flink's Watermark mechanism in the Kafka Source. As more and more operators get native support, we plan to merge connected native operators together for execution to reduce the R2C (row-to-columnar) conversion overhead. However, Auron's current Watermark mechanism isn't designed for this scenario, so we need to design a solution that fits this use case while still satisfying the Watermark constraints.

[Tartarus0zm]

Here's a half-baked idea I have:

• we could introduce a dedicated flink_watermark_exec.rs specifically for generating Flink Watermarks, with Kafka partition factors taken into account internally. Then we could extend RecordBatch by introducing meta fields to handle the Watermark mechanism.

• Within Flink native operators, we'd use this format uniformly. When sending data to Flink Java operators, on one hand we'd do the C2R conversion, and on the other hand we'd parse out the Watermark and forward it to the downstream operators.

If anyone has other ideas, feel free to discuss.