Prune values across the query executor instead of on the calling thread

[xiangfu0]

Summary

Value-based segment pruning uses the query executor for lists larger than 10 segments, bounded by the query's execution-thread limit. Small lists and calls without an executor retain serial execution.

ValueBasedSegmentPruner owns the shared serial/parallel loops, task-local caches, deadline handling, interruption checks, and per-segment fetch-context acquisition/release. BloomFilterSegmentPruner delegates to that implementation while retaining its prefetch planning and cleanup. Duplicate execution code and stale threshold comments are removed. Parallel results may have a different order.

Validation

• Built and tested against Apache master, independently of the DATA-3221 heap stack.
• 46 tests passed across ColumnValueSegmentPrunerTest, BloomFilterSegmentPrunerTest, SegmentPrunerServiceTest, and QueryMultiThreadingUtilsTest.
• Covers selected-set equivalence, executor use, serial fallback, expired deadlines, worker failures, interruption between segments, and Bloom fetch-context cleanup on success, failure, and cancellation.
• Spotless, Checkstyle, license formatting, and license checks passed for pinot-core.
• The additional strict compiler-warning build stopped in unchanged ZstandardDecompressor.java:51 because the Zstandard dependency references an unavailable JetBrains NotNull annotation. Isolated compilation reproduced this on the exact master source; normal compilation succeeds, and supplying the annotation jar also makes strict compilation succeed.