Summary

Fixes #2208

cleanup_old() ran two DELETE ... WHERE timestamp < ? scans on every record() and record_parse_failure() call, adding ~7ms of synchronous SQLite I/O to every RTK command — violating the <10ms startup target.

This PR gates cleanup behind a 24-hour interval using a new meta key-value table:

• On each record() call, a single SELECT on the 1-row meta table checks last_cleanup (~0.05ms)
• Cleanup only runs if >24 hours have elapsed since the last run
• 90-day data retention means daily cleanup is more than sufficient

Performance impact:

• Before: ~7ms added to every command (two DELETE table scans)
• After: ~0.05ms on 99.9% of calls (one SELECT on PK), ~7ms once per 24h

Changes:

• Add CLEANUP_INTERVAL_HOURS constant (24h) to constants.rs
• Add meta table (key TEXT PRIMARY KEY, value TEXT) in schema setup
• Replace cleanup_old() with maybe_cleanup() — checks last_cleanup timestamp before running DELETEs
• Use let _ = for cleanup calls (INSERT already succeeded; cleanup failure shouldn't break recording)

Test plan

• cargo fmt --all — no issues
• cargo clippy --all-targets — zero warnings
• cargo test --all — all 2246 tests pass
• New test: test_maybe_cleanup_skips_when_recent — verifies old records survive when last_cleanup is fresh
• New test: test_maybe_cleanup_runs_when_stale — verifies old records are deleted when last_cleanup >24h ago