[FLINK-39758] Fix DECIMAL OOB in SchemaMergingUtils

[haruki-830]

Summary

This commit fixes the DECIMAL precision overflow (Out-Of-Bounds) issue in SchemaMergingUtils and SchemaUtils, which previously caused pipeline failures when merging DECIMAL types from multiple tables with significantly different precisions and scales.

Key Changes

1. New createDecimalBounded Utility Method

• Introduced createDecimalBounded method in SchemaMergingUtils to handle DECIMAL precision overflow gracefully
• When computed precision exceeds DecimalType.MAX_PRECISION (38), the method truncates scale digits instead of throwing an exception
• Truncation formula: lossDigits = precision - MAX_PRECISION, then resultPrecision = precision - lossDigits, resultScale = scale - lossDigits
• Follows the same approach as Flink SQL for handling DECIMAL overflow

2. SchemaMergingUtils Fix

• Replaced Preconditions.checkArgument with createDecimalBounded call in mergeDecimalTypes()
• Removed the hard failure that threw IllegalArgumentException when merged precision exceeded 38
• SchemaUtils Fix
• Applied the same createDecimalBounded fix in inferWiderType() for DECIMAL-to-DECIMAL merging
• Applied the same fix in mergeExactNumericsIntoDecimal() for DECIMAL-to-integer merging

3. Comprehensive Testing

• Updated SchemaUtilsTest to verify that overflow cases now return bounded DECIMAL types instead of throwing exceptions
• Added testMergingDecimalWithOutOfBoundPrecisions with 6 parameterized test cases covering various overflow scenarios
• Refactored FlinkPipelineComposerITCase to support generic merging test infrastructure via runGenericMergingTest

Before fix: Merging DECIMAL(5, 5) and DECIMAL(38, 0) throws IllegalArgumentException (requires precision 43, max 38)
After fix: Merging DECIMAL(5, 5) and DECIMAL(38, 0) returns DECIMAL(38, 0)

JIRA Reference

https://issues.apache.org/jira/browse/FLINK-39758

[haruki-830]

@lvyanquan could you please help review this PR