[stanbrub]

DH-23641 (deephaven-core #8878) decodes only the requested rows when a Parquet read touches a small fraction of a page. No nightly or release benchmark exercises that path: standard tests select() their source before timing, and the ParquetRead tests time full reads. This adds one sparse benchmark and a dense control to ParquetMultiColTest, so a later change can't regress the path unnoticed.

• FileTestRunner.runParquetFilteredReadTest: reads the file, filters on long10K, then selects str10K, int10K, short10K and bigDec10K for the surviving rows. The rate is file rows per second, like the other read tests.
• ParquetRead- Sparse Multi Col -Static keeps 1 in 100 rows; ParquetRead- Dense Multi Col -Static keeps 1 in 2. The PR reads a page sparsely when under 12.5% of it is requested. The file's pages hold at least 5,184 rows for these columns, so the two tests fall clearly on either side of that cutoff.
• The array columns are left out because the PR doesn't read repeated columns sparsely; they would only dilute the result.
• Both tests read the Snappy Defaults file written earlier in the class, and they run in the nightly with the rest of ParquetMultiColTest.

Test Completed

• B/A on the nightly server: main b7ed9cb7c8 vs #8878 243a7f5b8e, 90M rows, 5 runs each, alternating build order.
• Every #8878 run beat every main run on the sparse test (6.91 s → 3.16 s).
• On main, keeping 1% of rows costs as much as keeping 50%, because pages are decoded whole either way.
• The existing reads and writes in the class showed no consistent change.

[Copilot]

🟢 Approval recommended

No active review issues are identified for the sparse and dense Parquet read benchmark changes.

No open issues 🎉

What changed in this PR

Adds sparse and dense Parquet read benchmarks to detect regressions in selective row decoding, using the existing ParquetMultiColTest benchmark suite.

Changes:

• Adds a filtered-read helper that filters on long10K, selects requested columns, and reports throughput in file rows per second.
• Adds 1% sparse and 50% dense reads of the existing Snappy Defaults file, excluding array columns.
• Makes scale-factor setup explicit per test and reduces the Gzip tests’ scale factor.

File	Description
src/​it/​java/​io/​deephaven/​benchmark/​tests/​standard/​file/​ParquetMultiColTest.java	Adds sparse/dense benchmark cases and per-test scale-factor setup.
src/​it/​java/​io/​deephaven/​benchmark/​tests/​standard/​file/​FileTestRunner.java	Adds the reusable filtered Parquet read benchmark helper.

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No active review issues remain for the filtered Parquet benchmark changes.

No open issues 🎉

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No active review issues are reported for the Parquet read benchmark changes.

No open issues 🎉

_{Review effort: Balanced}