Optimize tag index timeseries lookup

[Caideyipi]

Summary

• use direct map lookup for exact tag-value predicates instead of scanning every distinct value
• apply the path pattern before sorting matched measurements
• add a regression test that verifies direct lookup, path filtering, and result ordering

Test

• mvn test -pl iotdb-core/datanode -Dtest=TagManagerTest -DfailIfNoTests=false -Dcheckstyle.skip=true
• mvn spotless:apply -pl iotdb-core/datanode

Manual performance UT

TagManagerPerformanceTest is skipped by default. Enable it in a normal Maven build with:

mvn test -pl iotdb-core/datanode -Dtest=TagManagerPerformanceTest -Diotdb.tag.index.perf.enabled=true -DfailIfNoTests=false

Local index-lookup microbenchmark on Windows 11, Intel Core i9-13900H, and Java HotSpot 17.0.15. The legacy side reproduces the previous tag-value scan, sort, then path filter; the optimized side calls the current TagManager lookup. The test asserts identical ordered results. Values are medians of 5 rounds with 30 warmups: 10,000 distinct tag values (1 exact match), 2,000 candidate nodes (100 path matches), and 50,000 exact-lookup or 1,000 path-filter iterations per round.

Scenario	Legacy CPU (us/query)	Optimized CPU (us/query)	CPU speedup	Legacy allocation (B/query)	Optimized allocation (B/query)	Allocation change
Exact value lookup	63.750	1.250	51.00x	760	872	+14.7%
Path filter before sort	1,609.375	828.125	1.94x	337,720	291,688	-13.6%

Local validation: the default-disabled invocation skipped the benchmark as intended; the enabled test passed; TagManagerTest passed all 4 tests; Spotless and Checkstyle passed.

Build note: full DataNode Maven test-compile failed locally in unrelated main-source/generated fill and mode classes. For these measurements, the current TagManager.java and performance test were compiled with javac against cached 2.0.11 artifacts and run through JUnitCore. The numbers cover index lookup, not an end-to-end query.

[jt2594838]

May add a performance comparison