feat(core): support input and lookup changelog producers

[lszskye]

Purpose

Linked issue: #174

Support input and lookup changelog producers for primary-key tables.

This change:

• Writes raw input changelog records into changelog files while write process.
• Rolls input changelog files by target file size, consistently with Java Paimon.
• Produces lookup changelogs during compaction by comparing level-0 records with existing records.
• Supports lookup changelog with:
  • several merge engines;
  • deletion vectors enabled;
  • ignore_previous_files;
  • row deduplication and ignored comparison fields;
  • schema evolution and external data paths.
• Persists keys in first-row lookup SST files so excluded higher-level records can be detected.
• Supports shared shredding when writing input and compaction changelog files.
• Supports configurable changelog file prefix, format, and compression.
• Reads changelog manifests during streaming scans.
• Excludes level-0 files from the initial full scan for lookup changelog tables to avoid duplicate emission.
• Adds order-independent equality comparison for MAP values.
• Keeps full-compaction changelog producer unsupported.

Tests

Added and updated unit and integration coverage for:

• input and lookup changelog stream reads;
• insert, update, and delete changelogs;
• first-row and deletion-vector lookup;
• row deduplication and ignored fields;
• schema evolution and external paths;
• configurable changelog format, compression, and prefix;
• shared-shredding changelog files;
• nested row and order-independent MAP equality;
• changelog manifest scanning.

API and Format

This change adds public option constants for:

• changelog-producer.row-deduplicate
• changelog-producer.row-deduplicate-ignore-fields
• changelog-file.prefix
• changelog-file.format
• changelog-file.compression

[lxy-9602]

+1