sync migration dry-run reports wrong verb (inserted vs updated)

[moekatib]

Summary

sync run --dry-run on a legacy .one/sync/data/<plat>.db migration predicts inserted: 30 on rows that a real run would updated. Cosmetic only — no data impact — but misleading.

Repro

1. Have a legacy SQLite DB at .one/sync/data/fathom.db with rows that already exist in the memory store under the same source keys.
2. one --agent sync run fathom --dry-run
3. Output reports inserted: N where every row would hit the upsert-by-keys updated branch on a real run.

Expected

Dry-run runs the same classify-before-upsert path as a live run (or at least tests mem_records.keys && p_keys against the current set) so the predicted action matches.

Notes

• Not blocking. Most users don't dry-run migrations.
• Found during verification of PR feat: unified memory (v1.42.0) #125 (--full-refresh reconcile gate + upsert self-heal).

[moekatib]

Fixed in de8a302: migrate dry-run now probes the backend with a SELECT 1 FROM mem_records WHERE keys && $1 LIMIT 1 before incrementing the counter, so the predicted verb (inserted vs updated) matches what a live run would do. Requires backends with rawSql: true (both first-party plugins).