feat(notify): per-channel delivery results + accurate notification metrics

Dispatcher.Send now returns []ChannelResult{Type, Err} alongside the
aggregated error, and scheduler.checkDomain increments
NotificationsTotal per channel type/status instead of a single
unconditional IncNotification("dispatch", newStatus) placeholder that
ignored per-channel delivery outcome.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BwxdSt4reTm7Dj1oxRvpP3

internal/scheduler/scheduler.go

@@ -49,7 +49,7 @@ type Checker interface {
 // NotifySender delivers a status-change event to a project's notification
 // channels. internal/notify.Dispatcher satisfies this.
 type NotifySender interface {
-	Send(ctx context.Context, projectID uuid.UUID, ev notify.Event) error
+	Send(ctx context.Context, projectID uuid.UUID, ev notify.Event) ([]notify.ChannelResult, error)
 }
 
 // Scheduler drives periodic domain checks for every due project schedule.
@@ -172,10 +172,17 @@ func (s *Scheduler) checkDomain(ctx context.Context, projectID uuid.UUID, d stor
 			Summary: summarize(newStatus, cs, checkErr),
 			At:      now,
 		}
-		if err := s.notifier.Send(ctx, projectID, ev); err != nil {
+		results, err := s.notifier.Send(ctx, projectID, ev)
+		if err != nil {
 			log.Printf("scheduler: notify send for project %s domain %s failed: %v", projectID, d.ID, err)
 		}
-		s.metrics.IncNotification("dispatch", newStatus)
+		for _, r := range results {
+			status := "sent"
+			if r.Err != nil {
+				status = "failed"
+			}
+			s.metrics.IncNotification(r.Type, status)
+		}
 	}
 
 	return newStatus