Group small columns together in parquet files

[raunaqmorarka]

Description

Modified parquet writer to store columns in order of their size inside row groups
so that the reader can fetch small columns in fewer filesystem requests

Additional context and related issues

Based on similar logic in ORC writer at

trino/lib/trino-orc/src/main/java/io/trino/orc/OrcWriter.java

Line 420 in a13e451

Collections.sort(dataStreams);

Release notes

( ) This is not user-visible or docs only and no release notes are required.
( ) Release notes are required, please propose a release note for me.
(x) Release notes are required, with the following suggested text:

# Hive
* Improve the layout of data in parquet files produced by the optimized parquet writer for faster reads. ({issue}`17404`)

# Hudi, Iceberg, Delta
* Improve the layout of data in parquet files for faster reads. ({issue}`17404`)

[sopel39]

Do you think it might drop performance in some scenarios?

E.g. it seems optimally it should group columns that are mostly used together

[raunaqmorarka]

Do you think it might drop performance in some scenarios?

E.g. it seems optimally it should group columns that are mostly used together

This change matters only for columns with size less than parquet.max-buffer-size (default 8mb). Given that we're only affecting smaller columns, the penalty for a bad decision shouldn't be high and the current heuristic is the best we can do without knowing usage patterns. I'm also relying on this having been a successful optimization (or at least no complaints) in ORC.