T3.2: Add app event types and observer dispatch

Add src/app/events.go: a sealed Event interface with three concrete
types (JobChanged, RunRecorded, SchedulerStateChanged), an Observer
interface plus ObserverFunc adapter, and Subscribe/emit on the Service.
This replaces the scheduler's single onChange callback with typed
events the UI can exhaustively type-switch over.

Dispatch is serialized by a dedicated dispatchMu (separate from the
state lock): observers never run concurrently, emit must be called
without holding s.mu so observers can read Service state, and observers
must not re-enter an emitting method. emit is wired to mutating ops in
T3.3. Adds tests for ordered multi-observer delivery, empty-observer
no-op, and observer-reads-state-without-deadlock.

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

src/app/events.go

@@ -0,0 +1,91 @@
+package app
+
+import "gitea.mixdep.ru/mix/gosentry/src/domain"
+
+// Event is something the Service did to its state that observers may want to
+// react to. It is a sealed interface: the concrete types in this file are the
+// only implementations (enforced by the unexported isEvent marker), so a UI
+// listener can exhaustively type-switch over them and the compiler will flag a
+// new event type that a switch forgot to handle.
+//
+// Events replace the old single onChange callback. Instead of the scheduler
+// reaching into the GUI, the Service emits typed events and the UI subscribes —
+// the UI's listener becomes the one place that touches widgets.
+type Event interface {
+	isEvent()
+}
+
+// JobChanged signals that a job's durable config or transient runtime changed:
+// created, edited, deleted, enabled/disabled, or a status transition such as a
+// run starting. Observers should re-read the affected state through the Service
+// (Jobs/Runtime) rather than expect a payload snapshot — that keeps the event
+// small and avoids handing out stale copies.
+//
+// JobID identifies the affected job. A zero JobID means a broad change (for
+// example a delete, or a global pause that touched every job) and observers
+// should refresh their whole view.
+type JobChanged struct {
+	JobID int
+}
+
+// RunRecorded signals that a job run finished and produced a RunRecord. It
+// carries the record by value because the record is an immutable result that
+// observers append to history; there is nothing for them to re-read.
+type RunRecorded struct {
+	Record domain.RunRecord
+}
+
+// SchedulerStateChanged signals that the global scheduler pause state flipped.
+// The UI uses it to update the pause/resume control and status text.
+type SchedulerStateChanged struct {
+	Paused bool
+}
+
+func (JobChanged) isEvent()            {}
+func (RunRecorded) isEvent()           {}
+func (SchedulerStateChanged) isEvent() {}
+
+// Observer receives events emitted by the Service. OnEvent is the single
+// reaction point; the UI implements it and marshals any widget work onto the
+// main thread (fyne.Do) itself — the Service knows nothing about Fyne.
+type Observer interface {
+	OnEvent(Event)
+}
+
+// ObserverFunc adapts a plain function to the Observer interface, so callers can
+// subscribe a closure without declaring a type.
+type ObserverFunc func(Event)
+
+// OnEvent calls the wrapped function.
+func (f ObserverFunc) OnEvent(event Event) { f(event) }
+
+// Subscribe registers an observer to receive every subsequently emitted event.
+// Registration is expected during setup, before the scheduler starts, but is
+// guarded so it is safe at any time.
+func (s *Service) Subscribe(observer Observer) {
+	s.dispatchMu.Lock()
+	defer s.dispatchMu.Unlock()
+	s.observers = append(s.observers, observer)
+}
+
+// emit delivers an event to every registered observer.
+//
+// Single-threaded dispatch contract:
+//   - emit holds dispatchMu for the whole dispatch, so observers are never
+//     invoked concurrently and never overlap with each other or with Subscribe.
+//     Each observer sees events one at a time, in emit order.
+//   - emit must be called WITHOUT holding s.mu. The Service computes a state
+//     change under mu, releases it, then emits — so an observer is free to call
+//     back into read methods (Jobs/Runtime) without deadlocking on the state
+//     lock.
+//   - An observer must NOT call back into a Service method that emits (directly
+//     or indirectly): dispatchMu is non-reentrant, so re-entrant emission would
+//     deadlock. Observers react and return quickly; long or UI work is the
+//     observer's own responsibility to defer (e.g. fyne.Do).
+func (s *Service) emit(event Event) {
+	s.dispatchMu.Lock()
+	defer s.dispatchMu.Unlock()
+	for _, observer := range s.observers {
+		observer.OnEvent(event)
+	}
+}