feat(agent): event-driven uplink — sync on every state transition

The agent reported its state only on the adaptive sync tick (3s watching /
10s idle), so a resolving→downloading→verifying→organizing→completed
transition could lag up to a full interval before the server (and the web
UI) saw it. Now every successful Task.Transition fires an onChange hook
wired to TriggerSync, pushing the new state immediately. Bursts are safe:
TriggerSync is a buffered-1 send, so clustered transitions coalesce into
one sync.

- Task gains an onChange hook fired AFTER the status mutex is released
  (so a future heavier hook can't deadlock on task.mu); nil is a no-op.
- Manager.OnStateChange is set on each task at Submit; the daemon wires it
  to TriggerSync alongside the existing OnTaskDone.
- Stream tasks transition outside the Manager, so handleStreamTask wires
  the same hook explicitly (gap found in review) — resolving/downloading/
  completed/failed on the stream path now push too.

The adaptive ticker stays as a reconciliation heartbeat; it's just no
longer the latency floor for state changes.

internal/engine/manager.go

@@ -34,6 +34,13 @@ type Manager struct {
 	// Used by the daemon to trigger an immediate sync.
 	OnTaskDone func()
 
+	// OnStateChange is called after EVERY successful task status transition
+	// (resolving → downloading → verifying → organizing → seeding → done/failed),
+	// wired by the daemon to trigger an immediate sync so the server sees state
+	// changes in near-realtime instead of on the next adaptive tick. Coalesced
+	// downstream (TriggerSync is a buffered-1 send), so bursts collapse safely.
+	OnStateChange func()
+
 	// recentlyFinished holds tasks that completed/failed since the last sync read.
 	// The sync goroutine reads and clears this to include final states in the next sync.
 	recentMu       sync.Mutex
@@ -82,6 +89,8 @@ func NewManager(cfg ManagerConfig, reporter *ProgressReporter, downloaders ...Do
 // Submit queues a task for download. Non-blocking if capacity available.
 func (m *Manager) Submit(ctx context.Context, at agent.Task) {
 	task := NewTaskFromAgent(at)
+	// Event-driven uplink: push every status transition to the server immediately.
+	task.SetOnChange(m.OnStateChange)
 
 	// Per-task cancellable context so CancelTask can unblock the goroutine
 	taskCtx, taskCancel := context.WithCancel(ctx)

internal/engine/task.go

@@ -78,6 +78,12 @@ type Task struct {
 	ClaimedAt   time.Time
 	StartedAt   time.Time
 	CompletedAt time.Time
+
+	// onChange, when set, is called after every successful status Transition so
+	// the daemon can push the new state to the server immediately (event-driven
+	// uplink) instead of waiting for the next sync tick. Must be non-blocking —
+	// it's a coalescing TriggerSync. Set by the Manager at submit time.
+	onChange func()
 }
 
 // NewTaskFromAgent creates a Task from a server-claimed agent.Task.
@@ -111,13 +117,15 @@ func NewTaskFromAgent(at agent.Task) *Task {
 	}
 }
 
-// Transition validates and performs a state transition.
+// Transition validates and performs a state transition. On success it invokes
+// the onChange hook (outside the lock) so the daemon can push the new state to
+// the server immediately rather than waiting for the next sync tick.
 func (t *Task) Transition(to TaskStatus) error {
 	t.mu.Lock()
-	defer t.mu.Unlock()
 
 	allowed, ok := validTransitions[t.Status]
 	if !ok {
+		t.mu.Unlock()
 		return fmt.Errorf("no transitions from %s", t.Status)
 	}
 	for _, a := range allowed {
@@ -129,12 +137,28 @@ func (t *Task) Transition(to TaskStatus) error {
 			if to == StatusCompleted || to == StatusFailed {
 				t.CompletedAt = time.Now()
 			}
+			cb := t.onChange
+			t.mu.Unlock()
+			// Fire the event-driven uplink AFTER releasing the lock so a future
+			// heavier hook can't deadlock on the task mutex.
+			if cb != nil {
+				cb()
+			}
 			return nil
 		}
 	}
+	t.mu.Unlock()
 	return fmt.Errorf("invalid transition: %s -> %s", t.Status, to)
 }
 
+// SetOnChange wires the post-transition hook. Call before the task starts
+// transitioning (the Manager sets it at submit time).
+func (t *Task) SetOnChange(fn func()) {
+	t.mu.Lock()
+	t.onChange = fn
+	t.mu.Unlock()
+}
+
 // GetStatus returns current status thread-safely.
 func (t *Task) GetStatus() TaskStatus {
 	t.mu.RLock()

internal/engine/task_test.go

@@ -216,3 +216,41 @@ func TestHasUntried(t *testing.T) {
 		t.Error("all methods tried")
 	}
 }
+
+func TestTransitionFiresOnChange(t *testing.T) {
+	task := NewTaskFromAgent(agent.Task{ID: "t1"}) // StatusClaimed
+
+	var fired int
+	task.SetOnChange(func() { fired++ })
+
+	// Valid transition fires the hook.
+	if err := task.Transition(StatusResolving); err != nil {
+		t.Fatalf("Transition: %v", err)
+	}
+	if fired != 1 {
+		t.Errorf("onChange fired %d times, want 1 after a valid transition", fired)
+	}
+
+	// Another valid transition fires again (event-driven, every transition).
+	if err := task.Transition(StatusDownloading); err != nil {
+		t.Fatalf("Transition: %v", err)
+	}
+	if fired != 2 {
+		t.Errorf("onChange fired %d times, want 2", fired)
+	}
+
+	// Invalid transition must NOT fire the hook.
+	if err := task.Transition(StatusClaimed); err == nil {
+		t.Error("expected error on invalid transition downloading→claimed")
+	}
+	if fired != 2 {
+		t.Errorf("onChange fired %d times, want still 2 (no fire on invalid transition)", fired)
+	}
+}
+
+func TestTransitionNilOnChangeNoPanic(t *testing.T) {
+	task := NewTaskFromAgent(agent.Task{ID: "t2"}) // no onChange set
+	if err := task.Transition(StatusResolving); err != nil {
+		t.Fatalf("Transition with nil onChange must not error: %v", err)
+	}
+}