feat: migrate on-device Qwen3 ASR to CoreML for background Flow dictation

Replace MLX GPU inference with CoreML bundles so transcription continues
while the host app is backgrounded. Adds model download and warm-up,
vendored Qwen3Speech, and updates onboarding, settings, and copy for the
~1.6 GB CoreML package (iOS 18+).

OSGKeyboardShared/Services/FlowSessionKeys.swift

@@ -13,17 +13,39 @@ public enum FlowSessionKeys {
     public static let keyboardRecordingState = "flow.keyboardRecordingState"
     public static let transcriptionLanguage = "flow.transcriptionLanguage"
     public static let transcriptionResult = "flow.transcriptionResult"
+    /// Soft warning when polish failed but raw transcript was delivered.
+    public static let transcriptionPolishWarning = "flow.transcriptionPolishWarning"
     public static let transcriptionError = "flow.transcriptionError"
     public static let audioLevels = "flow.audioLevels"
 
-    /// Heartbeat older than this implies the host app was killed.
+    /// Heartbeat older than this while the host is foreground → likely killed.
     public static let heartbeatStaleInterval: TimeInterval = 3
 
     /// Default Flow session length when started from the keyboard.
     public static let defaultSessionDuration: TimeInterval = 480
 
-    /// Maximum duration for a single keyboard utterance.
-    public static let maxUtteranceDuration: TimeInterval = 60
+    /// Maximum duration for a single keyboard utterance (3 minutes).
+    public static let maxUtteranceDuration: TimeInterval = 180
+
+    /// Host polls for pipelined ASR drain after mic stop. Pipelining usually
+    /// finishes most chunks during recording; this is a soft deadline before
+    /// blocking on `asrTask.value` (which waits until the pipeline exits).
+    public static let localASRWaitTimeout: TimeInterval = 120
+    public static let localQwen3ASRWaitTimeout: TimeInterval = 180
+    public static let cloudASRWaitTimeout: TimeInterval = 120
+
+    /// Keyboard watchdog after the user stops recording (not utterance max length).
+    /// Must cover worst-case post-stop backlog: remaining MLX/SpeechAnalyzer chunks
+    /// plus cloud LLM polish (see `PolishingService.effectiveTimeout` cap).
+    public static func keyboardResultTimeout(
+        engineMode: String,
+        localASRBackend: LocalASRBackend
+    ) -> TimeInterval {
+        if engineMode == "local" {
+            return localASRBackend == .qwen3ASR ? 240 : 180
+        }
+        return 240
+    }
 
     public enum RecordingState: String, Sendable, Equatable {
         case idle