test: add ut for release-0.3

[lszskye]

Purpose

This change improves reader correctness and expands coverage for several previously untested paths.

• Make asynchronous union global-index tasks own their actions and predicate arguments, avoiding dangling references when an earlier task throws while other tasks remain queued.
• Align streaming $binlog behavior by packing adjacent UPDATE_BEFORE and UPDATE_AFTER records into one list-valued update event.
• Avoid adding _VALUE_KIND twice when constructing no-merge read schemas.
• Fix $files statistics projection for partitioned partial writes by mapping statistics only to columns physically written to the data file.
• Add coverage for bucket selection, Hive smallint hashing, list aggregation, compaction rewrites, eager deletion-file generation, index partition filtering, and manifest cleanup after failures.

Tests

• deferred asynchronous union-index tasks after an earlier task throws
• streaming binlog update packing within and across batches
• unmatched UPDATE_BEFORE handling
• binlog sequence-number schema
• partitioned partial-write statistics in $files
• default, Hive, and MOD bucket selection
• Hive smallint bucket compatibility
• listagg distinct mode with an empty delimiter
• full-compaction file rewrites
• eager deletion-vector generation
• partition-filtered index manifest reads
• cleanup of newly generated manifest files after merge failure

[lxy-9602]

+1