Update flow diagrams for event bus architecture, cancel cleanup, and SubscribeEvents

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/claude-agent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

docs/architecture/flows/event-bus-architecture.puml

@@ -0,0 +1,86 @@
+@startuml Event Bus Architecture
+skinparam componentAlign center
+title Event Bus: In-Process Pub/Sub Architecture
+
+package "Publishers" {
+    [Workflow Goroutine 1\n(album A, LOSSLESS)] as WF1
+    [Workflow Goroutine 2\n(album B, LOSSY)] as WF2
+}
+
+database "PostgreSQL" as DB {
+    [workflow_runs] as WR
+    [album_events] as AE
+}
+
+package "Event Bus (in-memory)" {
+    [Topic: albumA:LOSSLESS] as T1
+    [Topic: albumB:LOSSY] as T2
+    [Global Subscribers] as GS
+}
+
+package "Subscribers" {
+    [MonitorAlbumStream\nClient A (album A)] as S1
+    [MonitorAlbumStream\nClient B (album A)] as S2
+    [SubscribeEvents\nClient C (global)] as S3
+}
+
+WF1 --> DB : 1. Write event\n(synchronous)
+WF1 --> T1 : 2. Publish\n(async notification)
+
+WF2 --> DB : 1. Write event
+WF2 --> T2 : 2. Publish
+
+T1 --> S1 : Ring buffer\n(per subscriber)
+T1 --> S2 : Ring buffer
+T1 --> GS
+T2 --> GS
+
+GS --> S3 : Ring buffer
+
+note right of DB
+  **Source of truth.**
+  Events survive restarts.
+  Replay via seq numbers.
+end note
+
+note right of T1
+  **Ephemeral notification.**
+  Ring buffer per subscriber.
+  Slow subscribers: overwrite oldest.
+  No backpressure on publishers.
+end note
+
+note bottom of S1
+  Client disconnect removes
+  subscriber from topic.
+  Workflow continues.
+end note
+
+== Subscription Lifecycle ==
+
+note as N1
+**Subscribe flow:**
+1. Client calls MonitorAlbumStream or SubscribeEvents
+2. Server subscribes to EventBus (per-topic or global)
+3. Server queries DB for historical events (replay)
+4. Server bridges: EventBus → gRPC stream
+5. On disconnect: cleanup func unsubscribes
+
+**Topic cleanup:**
+When last subscriber leaves AND workflow completed:
+topic removed from EventBus map.
+end note
+
+== Recovery on Restart ==
+
+note as N2
+**Server restart recovery:**
+1. Query workflow_runs WHERE status = 'running'
+2. For each stale run:
+   - If active download exists → mark completed
+   - Otherwise → mark failed ("server restarted")
+3. RecoverOrphanedDownloads reschedules poll jobs
+4. New workflows start fresh (no goroutine resurrection)
+end note
+
+@enduml