[Pipe] Batch OPC UA sink writes

[Caideyipi]

Description

Target master.

Cherry-pick #18380 from dev/1.3.

The OPC UA sink previously issued and awaited one writeValue RPC for every measurement in a tablet. This change collects the writes for a tablet and sends them through one writeValues request.

Missing-node recovery is preserved:

• inspect each returned status and collect only Bad_NodeIdUnknown writes;
• deduplicate shared nodes before calling addNodes;
• retry only the writes whose nodes were missing;
• continue to fail immediately on other bad statuses with measurement context.

The cherry-pick keeps the compile-time i18n messages and Tablet#setRowSize API used on master.

Unit tests cover a successful multi-measurement batch, missing-node creation and selective retry, and a non-recoverable write status.

Verification

mvn -o -nsu -pl iotdb-core/datanode \
  -Ddevelocity.off=true \
  -DskipTests \
  clean test-compile

mvn -o -nsu -pl iotdb-core/datanode \
  -Ddevelocity.off=true \
  -Dtest=IoTDBOpcUaClientTest \
  -Dsurefire.failIfNoSpecifiedTests=false \
  surefire:test

Results:

• module main and test compilation passed;
• 3 tests run, 0 failures, 0 errors, 0 skipped;
• Checkstyle and Spotless passed;
• git diff --check origin/master...HEAD passed.

Local synthetic benchmark

The local-only benchmark from #18380 compared the previous per-measurement request pattern with the batch request under a simulated 5 ms response delay:

Measurements per tablet	Previous RPCs	Batch RPCs	Previous median	Batch median	Speedup
100	100	1	1547.574 ms/tablet	15.329 ms/tablet	100.96x

This is a synthetic RPC-count and latency comparison rather than a measurement of production OPC UA server throughput.

---

This PR has:

• been self-reviewed.
• added unit tests for the new batch and recovery paths.

---

Key changed/added classes

• IoTDBOpcUaClient
• IoTDBOpcUaClientTest