Enable core Arrow codecs in s390x pyarrow build for datascience runtime

[coderabbitai]

Problem Description

The s390x pyarrow build in runtimes/datascience/ubi9-python-3.12/Dockerfile.cpu (lines 99-103) explicitly disables core Arrow codecs:

-DARROW_WITH_LZ4=OFF \
-DARROW_WITH_ZSTD=OFF \
-DARROW_WITH_SNAPPY=OFF \

This configuration prevents reading most real-world Parquet and Arrow datasets that use these common compression formats, significantly limiting the functionality of the datascience runtime on s390x architecture.

Impact Analysis

• Data Compatibility: Users cannot read Parquet files compressed with LZ4, Zstd, or Snappy (the most common compression formats)
• Runtime Failures: Applications attempting to read compressed datasets will fail with codec-related errors
• User Experience: s390x datascience runtime becomes significantly less capable than other architectures

Root Cause

The codecs were likely disabled to avoid build complexity or dependencies, but with -DARROW_DEPENDENCY_SOURCE=BUNDLED already set, the required codec libraries should be built in-tree without requiring additional system dependencies.

Solution

Enable the core codecs in the Arrow build configuration:

# Change from:
-DARROW_WITH_LZ4=OFF \
-DARROW_WITH_ZSTD=OFF \
-DARROW_WITH_SNAPPY=OFF \

# To:
-DARROW_WITH_LZ4=ON \
-DARROW_WITH_ZSTD=ON \
-DARROW_WITH_SNAPPY=ON \

Acceptance Criteria

• Core Arrow codecs (LZ4, Zstd, Snappy) are enabled in s390x pyarrow build
• s390x datascience runtime can successfully read Parquet files compressed with these formats
• Build time impact is acceptable (should be minimal with BUNDLED dependencies)
• No regression in build success rate for s390x architecture

Files Affected

• runtimes/datascience/ubi9-python-3.12/Dockerfile.cpu

Context

Identified during PR #1513 review: #1513 (comment)

The current implementation prioritizes build simplicity over runtime functionality, but enabling these codecs should not introduce significant complexity given the bundled dependency strategy.