fix(addons): dedup existing Ring devices before the unique index (R15)

Codex R15: on an upgrade where the pre-index race already produced duplicate
config.ring_id rows, CREATE UNIQUE INDEX would abort the migration. Now the
migration first collapses duplicates (keep the earliest id per ring_id; the
rest cascade-delete their entities/state) using portable subquery SQL that
runs on Postgres (prod) and SQLite (tests) alike.

Verified: 3 duplicates → 1 on Postgres, index then created; suite 43 green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

database/migrations/2026_07_18_060001_add_ring_id_unique_index_to_devices.php

@@ -7,6 +7,23 @@ return new class extends Migration
 {
     public function up(): void
     {
+        // Reconcile any duplicates the pre-index race may already have created, otherwise the
+        // CREATE UNIQUE INDEX below would abort on existing data. Keep the earliest row per
+        // ring_id; the rest cascade-delete their entities/state (both FKs cascadeOnDelete).
+        // Portable subquery form (works on Postgres in prod and SQLite in tests).
+        DB::statement(<<<'SQL'
+            DELETE FROM devices
+            WHERE config->>'ring_id' IS NOT NULL
+              AND id NOT IN (
+                  SELECT min_id FROM (
+                      SELECT MIN(id) AS min_id
+                      FROM devices
+                      WHERE config->>'ring_id' IS NOT NULL
+                      GROUP BY config->>'ring_id'
+                  ) keep
+              )
+        SQL);
+
         // Concurrent first-sight Ring messages (retained info + motion on bridge startup) must not
         // create two rows for one device. A partial unique index on the JSON ring_id lets the
         // second concurrent insert fail so the ingest job can re-fetch the winner (race-safe).