GH-32381: [C++] Improve error handling for hash table merges

[helloitsheqing]

Original Issue: #32381

Description

This pull request introduces a comprehensive test suite for the hash aggregation functionality in Apache Arrow’s compute kernels. The new test suite covers the following key components:

GroupedCountImpl:

• Validates merging of grouped counts with valid inputs.
• Tests behavior with invalid group ID mappings (e.g., mismatched sizes, out-of-bounds indices).

GroupedMinMaxImpl:

• Verifies correctness of merging minimum and maximum values across groups.
• Includes edge cases for invalid and out-of-bounds group ID mappings.

GroupedCountAllImpl:

• Tests merging of aggregated counts for "count all" scenarios.
• Handles cases with valid and invalid inputs, ensuring robust error handling.

The test suite is structured following the conventions of aggregate_test.cc, providing comprehensive coverage of valid and edge-case scenarios.

Key Changes

New Test File: Added hash_aggregate_test.cc in cpp/src/arrow/compute/kernels/.
CMake Configuration: Updated cpp/src/arrow/compute/kernels/CMakeLists.txt to include the new test file.
Test Coverage:
Each test focuses on verifying correctness, error handling, and robustness for the Merge methods of:

• GroupedCountImpl
• GroupedMinMaxImpl
• GroupedCountAllImpl

Future Work

If this test suite is approved, I plan to extend similar coverage to the remaining hash aggregation classes, including:

• GroupedTDigestImpl
• GroupedFirstLastImpl
• GroupedBooleanAggregator
• Other GroupedAggregator subclasses.
  This phased approach allows focused and iterative improvements to the test coverage for Arrow compute kernels.