Fix PBTree schema cache release after pre-delete

[Caideyipi]

Summary

• Make the PBTree test-only forceFlushAndRelease helper wait for outstanding flush/release workers before checking final memory statistics.
• Keep forcing release after all volatile buffers are flushed so low-cache PBTree statistics only retain the database root when expected.
• Release skipped pre-deleted measurements in Traverser#getChild so PBTree schema reads do not leak pinned cache entries.

Why the Traverser change matters

This is not only a test fix. In PBTree mode, fetchSchema and fetchSchemaWithoutWildcard set skipPreDeletedSchema=true to hide measurements that are in the pre-delete blacklist. For exact child lookups, store.getChild(...) pins the cached node before Traverser decides to skip it. The old code changed the child to null without unpinning it, leaving that measurement and its ancestors pinned. That can prevent eviction/release under real schema reads after pre-delete operations and inflate PBTree memory usage until the region is cleared.

Tests

• mvn spotless:apply -pl iotdb-core/datanode
• git diff --check
• mvn -pl iotdb-core/datanode -am -Dtest=SchemaStatisticsTest#testMemoryStatistics2 -Dsurefire.failIfNoSpecifiedTests=false -Dspotless.skip=true -Dcheckstyle.skip=true test -DtrimStackTrace=false fails before the target test runs during existing datanode compilation, with unrelated missing generated fill/aggregation symbols such as IFill, IFillFilter, Accumulator, and IoTDBConfig#getModeMapSizeThreshold().