[To dev/1.3] Optimize pipe tablet memory accounting

[Caideyipi]

Description

Backport #18211 to dev/1.3.

• Avoid creating an identity deduplication set for single InsertNode leaves and single-child composite nodes in InsertNodeMemoryEstimator.
• Adapt the single-raw-tablet string-intern-pool optimization to the Batch V1 request path used by dev/1.3.
• Preserve shared string interning for batches containing multiple raw tablets.
• Add opt-in manual performance UTs for the 100,000-column paths.

Tests

• mvn -o -pl iotdb-core/datanode spotless:apply
  • BUILD SUCCESS
• mvn -o -pl iotdb-core/datanode -Dcheckstyle.skip=true -Dspotless.skip=true -Dtest=InsertNodeMemoryEstimatorTest,PipeDataNodeThriftRequestTest test
  • Tests run: 25, Failures: 0, Errors: 0, Skipped: 0
• mvn -o -pl iotdb-core/datanode -Dcheckstyle.skip=true -Dspotless.skip=true -Dtest=InsertNodeMemoryEstimatorPerformanceTest,PipeTransferTabletBatchReqPerformanceTest -Diotdb.pipe.insert.node.memory.estimator.perf.enabled=true -Diotdb.pipe.insert.node.memory.estimator.perf.iterations=3 -Diotdb.pipe.insert.node.memory.estimator.perf.warmup.iterations=1 -Diotdb.pipe.tablet.batch.deserialize.perf.enabled=true -Diotdb.pipe.tablet.batch.deserialize.perf.iterations=3 -Diotdb.pipe.tablet.batch.deserialize.perf.warmup.iterations=1 test
  • Tests run: 2, Failures: 0, Errors: 0, Skipped: 0
  • InsertNode memory estimate benchmark: 100000 columns, 2.439-3.155 ms/op
  • Batch V1 single raw tablet deserialization benchmark: 100000 columns, 15.224-17.135 ms/op