[flink] Optimize sink statistics calculation to run only when necessary
Refactor RowDataSerializationSchema to extract size estimation logic into
RowDataSizeEstimator class and only enable it when PARTITION_DYNAMIC mode
is used. This avoids unnecessary computation overhead for other modes.
Other improvements:
- Improve operator names: "Collect Statistics", "Strip Statistics"
- Remove slotSharingGroup for "Strip Statistics" operator
- Remove redundant test setup code in Flink 1.18/1.19 IT cases
- Add comprehensive Javadoc for DistributionMode enum
- Add sink name with table path for better observability