Move independent Iceberg helpers to root-safe module [reduced-it]

[gerashegalov]

Related to #15821.

Supersedes #15826 with an independent, non-stacked change.

Description

Move a self-contained set of Apache Iceberg helpers into the root-safe
iceberg-common module without depending on the other pending Iceberg
unshimming pull requests.

The root-safe set includes:

• the deletion-vector container;
• reader option and native-row-index utilities;
• reader mode types; and
• position-delete field IDs.

Keep implementation-dependent delete-filter helpers in the versioned Iceberg
module by folding them into the existing GpuDeleteFileInfo companion. Remove
the six corresponding special root-layout allowlist entries and add the direct
spark-rapids-jni dependency required by the deletion-vector container.

The catalog and table wrappers remain in the versioned Iceberg module because
their Spark catalog interfaces are not source-compatible across all supported
Spark versions.

Packages, public class names, algorithms, data flow, and runtime behavior are
unchanged. This changes source-module ownership and class placement only.

This pull request was prepared with AI assistance and reviewed by the author
before publication. It also received an independent read-only code review.

TODO before marking ready for review:

• Confirm the full multi-shim distribution assembly in pre-merge CI.
• Confirm the AppClassLoader/extraClassPath Iceberg smoke coverage in
  pre-merge CI.
• Address CI and review feedback.

Validation:

• Built iceberg-common and the Iceberg consumer for Spark 3.5.0,
  Iceberg 1.6, and Scala 2.12.
• Built iceberg-common and the Iceberg consumer for Spark 4.1.1,
  Iceberg 1.11, and Scala 2.13.
• Built iceberg-common for Spark 3.3.1/Scala 2.12, Spark 4.0.0/Scala 2.13,
  and Spark 4.2.0/Scala 2.13 after keeping the catalog wrappers versioned.
• Ran GpuDeleteFilterSuite, GpuPostProcessorSuite,
  IcebergDeletionVectorSuite, and RapidsSparkTableSuite: 57 tests passed.
• Verified the expected classes are present in both root-safe classifier jars
  and absent from both versioned Iceberg jars.
• Ran ./build/make-scala-version-build-files.sh 2.13.
• Ran git diff --check.

A broader local reactor build stopped before reaching Iceberg because unchanged
SQL-plugin sources use a cuDF API marked deprecated by the locally resolved
dependency under warnings-as-errors. The focused changed modules and tests
completed successfully.

Performance: Not required because this only relocates helpers or changes their
owning companion. It does not add runtime operations, allocations, passes, or
synchronization.

Checklists

Documentation

• Updated for new or modified user-facing features or behaviors
• No user-facing change

Testing

• Added or modified tests to cover new code paths
• Covered by existing tests
  (GpuDeleteFilterSuite, GpuPostProcessorSuite,
  IcebergDeletionVectorSuite, and RapidsSparkTableSuite.)
• Not required

Performance

• Tests ran and results are added in the PR description
• Issue filed with a link in the PR description
• Not required

[greptile-apps]

Greptile Summary

The PR relocates self-contained Iceberg helpers into the root-safe iceberg-common module while retaining implementation-dependent helpers in the versioned module.

• Moves deletion-vector, Parquet reader utility, reader-mode, and position-delete field-ID classes.
• Updates consumers and removes obsolete root-layout allowlist entries.
• Consolidates delete-filter helpers into GpuDeleteFileInfo without changing their implementation.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
dist/unshimmed-common-from-single-shim.txt	Removes special root-layout entries for helpers now supplied by the root-safe module.
iceberg-common/src/main/java/com/nvidia/spark/rapids/iceberg/IcebergDeletionVector.java	Relocates the deletion-vector implementation without changing its algorithm or ownership behavior.
iceberg-common/src/main/scala/com/nvidia/spark/rapids/iceberg/parquet/GpuIcebergParquetReaderUtils.scala	Houses unchanged reader-option and native-row-index helpers in the shared module.
iceberg-common/src/main/scala/com/nvidia/spark/rapids/iceberg/parquet/ThreadConf.scala	Moves the root-safe reader-mode types out of the versioned implementation.
iceberg-common/src/main/scala/org/apache/iceberg/io/GpuPositionDeleteFieldIds.scala	Moves position-delete field IDs into a package-scoped shared helper.
iceberg/common/src/main/scala/com/nvidia/spark/rapids/iceberg/data/GpuDeleteFilter.scala	Consolidates unchanged delete-filter helpers into the existing GpuDeleteFileInfo companion.
iceberg/common/src/main/scala/com/nvidia/spark/rapids/iceberg/parquet/reader.scala	Replaces local helper definitions and calls with references to the root-safe utilities.
iceberg/common/src/main/scala/org/apache/iceberg/io/rolling.scala	Uses the relocated position-delete field-ID helper.
tests/src/test/spark350/scala/com/nvidia/spark/rapids/iceberg/GpuPostProcessorSuite.scala	Updates tests to invoke the relocated native-row-index utility.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Versioned[Versioned Iceberg module]
  Common[Root-safe iceberg-common]
  Dist[Distribution root]
  Runtime[Iceberg runtime consumers]
  Versioned -->|depends on| Common
  Common -->|promoted once| Dist
  Versioned --> Runtime
  Dist --> Runtime

Loading

_{Reviews (5): Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." | Re-trigger Greptile}

[gerashegalov]

build

[gerashegalov]

build

[NvTimLiu]

LGTM from CICD point of view

[res-life]

[P1] Update the remaining withNativeRowIndex callers

This PR removes object GpuIcebergParquetReader, but lines 1181 and 1218 in this file still call GpuIcebergParquetReader.withNativeRowIndex(...). A clean Spark 3.5 test compile therefore fails with not found: value GpuIcebergParquetReader. Please change those two callers to GpuIcebergParquetReaderUtils.withNativeRowIndex(...) as well.

[gerashegalov]

@res-life updated in 230ec0b

[res-life]

LGTM

[gerashegalov]

build