Pin the close before the reopen in the archive sweep (closes #103)

The archive sweep closes a target's archive connection before it reopens the file, and no test noticed if that close was removed, which would leak one SQLite connection per target per sweep. A test now keeps the connection from before a sweep that reopens the archive and requires it closed afterwards. The sweeper's query for database targets takes the sweep's context; a sweep whose context is done returns without an error line, so stopping is not logged as a failure, and a test pins that. The comment on the sweeper's cancel function gives the true reason it needs no lock: fx calls the stop hook only after the start hook has returned.

Model: opus-5-5

internal/delivery/archive_sweeper.go

@@ -45,8 +45,13 @@ type ArchiveSweeper struct {
 	eng      *Engine
 	log      *slog.Logger
 	interval time.Duration
-	cancel   context.CancelFunc
-	wg       sync.WaitGroup
+
+	// cancel needs no lock: fx calls the stop hook only after the
+	// start hook has returned, so stop never reads it while start
+	// is still setting it.
+	cancel context.CancelFunc
+
+	wg sync.WaitGroup
 }
 
 // NewArchiveSweeper creates the archive sweeper and registers
@@ -163,10 +168,18 @@ func (s *ArchiveSweeper) sweep(ctx context.Context) {
 	var targets []database.Target
 
 	err := s.db.DB().
+		WithContext(ctx).
 		Model(&database.Target{}).
 		Where("type = ?", database.TargetTypeDatabase).
 		Find(&targets).Error
 	if err != nil {
+		// The app stopping as a sweep starts cancels the listing.
+		// Stopping is not a failure, so it must not produce an
+		// error line.
+		if ctx.Err() != nil {
+			return
+		}
+
 		s.log.Error(
 			"archive sweep: failed to list database targets",
 			"error", err,