[rusq]

Large-channel database deduplication can select a channel/chunk range lookup that ignores message identity, making preview and cleanup impractically slow. Add a migrated MESSAGE (CHANNEL_ID, ID, CHUNK_ID) index so candidate searches constrain the message ID while preserving NULL-safe matching, edited payload history, and both dedupe modes. Existing archives build the index automatically on first open after upgrading.

Add migration, functional, and query-plan regression coverage plus separate Preview/Deduplicate benchmarks. Document the first-open migration cost and ignore trace files.

Validation:

• Focused database, dedupe-command, and resume race tests passed; all 98 non-scratch repository packages passed race/coverage tests and vet. Standard make targets encounter an existing compile error in ignored tmp/security/repro; the full suite was run with local listener access enabled.
• On a copy of a 620,723-message archive, CLI cleanup completed in 38.5 seconds, removing 25,543 messages, 310 files, and 267 chunks. Independent survivor comparisons, integrity/foreign-key checks, and a second no-op run passed. The original archive remained unchanged.
• Both message modes passed synthetic CLI checks. Two independent empty-context reviews reported no actionable findings.

Closes #750.