fix(tests): use portable bigint test literals

[ChaomingZhangCN]

Purpose

Linked issue: N/A

On macOS arm64, long and int64_t are distinct C++ types. As a result, expressions such as Literal(5l) instantiate Literal<long>, while Literal only explicitly instantiates its constructor for int64_t. This causes the Parquet test executable to fail during linking:

Undefined symbols for architecture arm64:
  "paimon::Literal::Literal<long>(long const&)"

This change replaces the plain long literals used by Parquet BIGINT predicate tests with an explicit int64_t helper. It follows the existing approach used by the corresponding ORC predicate tests.

Both predicate conversion and predicate pushdown tests are updated so that the complete paimon-parquet-format-test target can link successfully on macOS arm64.

Tests

Verified on macOS arm64:

cmake --build build --target paimon-parquet-format-test -j 8
./build/debug/paimon-parquet-format-test \
  --gtest_filter='PredicateConverterTest.*'
git diff --check HEAD^ HEAD

Results:

• paimon-parquet-format-test compiled and linked successfully.
• PredicateConverterTest.*: 7 tests passed.
• git diff --check: passed.

The complete Parquet test binary was also started locally. Some existing Parquet I/O tests fail during Arrow JSON test-data setup with std::bad_cast or an Arrow assertion. Those failures occur outside the code changed by this patch.

API and Format

No public API, storage format, or protocol changes.

Documentation

No documentation changes. This patch only improves test portability.

Generative AI tooling

Generated-by: OpenAI Codex (GPT-5)

[lxy-9602]

+1