Fix PathPatternNode wildcard cache lifecycle

[Caideyipi]

Description

PathPatternNode used to retain non-trivial wildcard child names and compile a regular expression on every match. This change caches the compiled Pattern objects instead.

• Compile each non-trivial wildcard pattern when the child is added.
• Reuse the compiled pattern in getMatchChildren.
• Remove cached patterns when children are deleted or cleared, including replacement of an existing wildcard child.
• Include the compiled-pattern cache in memory estimation.
• Reuse the same compilation helper from PathPatternUtil.isNodeMatch.

Testing

mvn spotless:apply -pl iotdb-core/node-commons
mvn -pl iotdb-core/node-commons -Dtest=PathPatternNodeTest test

Result: BUILD SUCCESS; 2 tests passed. Checkstyle and Spotless checks passed.

---

This PR has:

• been self-reviewed.
• added comments explaining the intent of the cache.
• added unit tests covering wildcard-cache add, replacement, deletion, and clear paths.

---

Key changed/added classes

• PathPatternNode
• PathPatternUtil
• PathPatternNodeTest

Local benchmark results

The follow-up benchmark from #18685 was run against the cached implementation from merge commit 9be80ead1dc and an in-test reproduction of the pre-change compile-on-every-match path. Environment: Windows 11 (build 26200), JDK 17.0.15, Intel Core i9-13900H, 2026-09-21. Each result is the median over 7 rounds after 2,000 warmup lookups; each round performed 20,000 lookups over 1,000 distinct queries.

mvn -pl iotdb-core/node-commons -Dtest=PathPatternNodePerformanceTest -Diotdb.path.pattern.node.perf.enabled=true -Diotdb.path.pattern.node.perf.wildcard-counts=1,10,100 -Diotdb.path.pattern.node.perf.query-count=1000 -Diotdb.path.pattern.node.perf.warmup-lookups=2000 -Diotdb.path.pattern.node.perf.lookups=20000 -Diotdb.path.pattern.node.perf.rounds=7 test

wildcard children	pre-change (ns/lookup)	cached (ns/lookup)	pre-change (lookups/s)	cached (lookups/s)	speedup	latency reduction
1	1,217.7	205.0	821,207	4,879,239	5.94x	83.17%
10	1,767.5	348.5	565,787	2,869,193	5.07x	80.28%
100	16,148.7	2,800.8	61,925	357,039	5.77x	82.66%

The pre-change column is deliberately implemented inside the benchmark for an apples-to-apples comparison; it is not a separately built old binary. The cached implementation is about 5-6x faster in this workload, with 80-83% lower lookup latency. The targeted Maven run passed (1 test, 0 failures; Checkstyle, Spotless, compilation, and Surefire all passed).