Parquet reader - avoid redundant dictionary reads

[ShubhamChaurasia]

Description

Avoid redundant dictionary reads - earlier the dictionary page is read at the time of rowgroup pruning (PredicateUtils) and then again read during the actual file reading(ParquetReader). This PR removes this redundant work and moves dictionary based rowgroup pruning logic to ParquetReader

Would like to thank @pettyjamesm for his initial suggestions on this.

Some numbers with patch

These queries were run on 1TB TPC-DS data(parquet, zstd compressed).

select count(*) from store_sales where ss_store_sk=163;

Without Patch - 26.14 [2.88B rows, 2.08GiB] [110M rows/s, 81.4MiB/s]
With Patch - 21.09 [2.88B rows, 1.09GiB] [137M rows/s, 52.9MiB/s]


select count(*) from store_sales where ss_hdemo_sk < 5226;

Without Patch - 22.23 [2.88B rows, 2.4GiB] [130M rows/s, 111MiB/s]
With Patch - 17.17 [2.88B rows, 1.25GiB] [168M rows/s, 74.6MiB/s]


select count(*) from store_sales where ss_store_sk=163 and ss_hdemo_sk < 5226;

Without Patch - 28.37 [2.88B rows, 5.07GiB] [102M rows/s, 183MiB/s]
With Patch - 18.98 [2.88B rows, 2.94GiB] [152M rows/s, 158MiB/s]


select ss_cdemo_sk, count(*) from store_sales where ss_store_sk=163 and ss_hdemo_sk < 5226 group by ss_cdemo_sk;

Without Patch - 28.87 [2.88B rows, 7.36GiB] [99.8M rows/s, 261MiB/s]
With Patch - 24.23 [2.88B rows, 5.23GiB] [119M rows/s, 221MiB/s]

Additional context and related issues

Release notes

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

## Section
* Fix some things. ({issue}`issuenumber`)

[starburstdata-automation]

Started benchmark workflow for this PR with test type = iceberg/sf1000_parquet_part.

Building Trino finished with status: success
Benchmark finished with status: failure
Comparing results to the static baseline values, follow above workflow link for more details/logs.
Status message: Found regressions for:<br/>(presto/tpcds, q09, totalCpuTime, over by 16.8%)<br/>(presto/tpcds, q88, totalCpuTime, over by 38.9%)
Benchmark Comparison to the closest run from Master: Report

[starburstdata-automation]

Started benchmark workflow for this PR with test type = iceberg/sf1000_parquet_unpart.

Building Trino finished with status: success
Benchmark finished with status: success
Comparing results to the static baseline values, follow above workflow link for more details/logs.
Status message: NO Regression found.
Benchmark Comparison to the closest run from Master: Report

[Praveen2112]

nit: Can we avoid null here ? Pass TupleDomain#all and an empty set ?

[pettyjamesm]

Nit: break each field onto a separate line

[pettyjamesm]

Do we really want to swallow an exception at this point? Seems like we should re-throw a variation of RuntimeException from dictionaryPredicateMatch (instead of throw IOException) and then not swallow that exception here since if we can't decode a dictionary page we should probably fail.