refactor: drop Qwen3 CoreML ASR, add local-engine cloud polish toggle

Rolls back the v0.2.0 Qwen3 CoreML on-device ASR stack and replaces the
'local engine' UX with iOS 26 SpeechAnalyzer + DictationTranscriber only.

The 'Cloud polish after ASR' toggle (ProviderConfig.localModeCloudPolishEnabled)
lets users opt into a post-ASR DeepSeek round-trip from the local engine.
Defaults to off so the local engine stays genuinely local. New PolishError.missingAPIError
surfaces an inline 'fill in your key' warning when the toggle is on but the
Keychain is empty. DeepSeek preset default model bumped to deepseek-v4-flash.

Deleted:
  - OSGKeyboard/ThirdParty/Qwen3Speech/ (74 files, ~16k LoC)
  - OSGKeyboard/Services/ModelManager.swift (492)
  - OSGKeyboard/Services/OnDeviceModelWarmup.swift (197)
  - OSGKeyboard/Services/Qwen3ASRService.swift (257)
  - OSGKeyboard/Services/ModelDownloadSourcePicker.swift (126)
  - OSGKeyboard/Views/OnDeviceModelsView.swift (184)
  - OSGKeyboard/Views/DownloadConfirmSheet.swift (96)
  - OSGKeyboardShared/Models/OnDeviceModel.swift (140)
  - OSGKeyboardShared/Services/OnDeviceModelStatus.swift (104)
  - Qwen3ASRServiceProvider registration in OSGKeyboardApp
  - Qwen3Speech package declaration in project.yml
  - 5 .qwen3ASR enum / branch reference sites in HomeView, OnboardingView,
    LocalEngineSettingsRows, FlowSessionManager, ASRService, EngineServiceLabel
  - Two pre-existing Swift 6 strict-concurrency errors in
    LiveDictationController + FlowSessionManager (the weak [weak self] in
    detached-task MainActor.run blocks) that were blocking clean builds

Added:
  - LocalModelsGroup: 'Built-in iOS SpeechAnalyzer' badge + 'Cloud polish
    after ASR' Switch toggle
  - PolishingService: honour localModeCloudPolishEnabled; new .missingAPIKey
    error case with localised warning
  - AppGroupStore.localModeCloudPolishEnabled (mirrored into App Group
    so the keyboard extension honours the toggle during live dictation)
  - SettingsView: show provider/api sections when local-mode cloud polish
    is on so the user can paste a DeepSeek key
  - FlowSessionManager: route through PolishingService for local + polish-on
    flow; translate missingAPIKey into a polished warning
  - KeyboardViewController: handle PolishingService.PolishError.missingAPIKey
    in the keyboard-side live polish path
  - CHANGELOG v0.2.1: documents the rollback + new toggle
  - README.md / README.zh.md: engine matrix section, data flow note

Verified: xcodebuild -scheme OSGKeyboard -destination 'generic/platform=iOS Simulator'
build succeeds under SWIFT_STRICT_CONCURRENCY=complete.

OSGKeyboardExt/KeyboardViewController.swift

@@ -565,6 +565,18 @@ public final class KeyboardViewController: UIInputViewController {
                     self.state.phase = .error(.llm(error), message: msg)
                     self.scheduleAutoClearError()
                 }
+            } catch let polishError as PolishingService.PolishError where polishError == .missingAPIKey {
+                // v0.2.0: local engine + cloud polish toggle on, but the
+                // user hasn't entered an API key. Insert the raw transcript
+                // (so the user doesn't lose what they said) and surface the
+                // same "fill in your key" hint we use in the cloud path.
+                self.textDocumentProxy.insertText(trimmed)
+                self.state.lastTranscript = ""
+                self.state.phase = .error(
+                    .llm(.noAPIKey),
+                    message: ExtL10n.string("keyboard.error.llm.noApiKey")
+                )
+                self.scheduleAutoClearError()
             } catch {
                 // Network / timeout / decoding — fall back to the raw
                 // transcript so the user still gets their text, with a

OSGKeyboardExt/Services/AppGroupPersistor.swift

@@ -35,10 +35,11 @@ public struct AppGroupPersistor {
         state.mode             = .polish
         state.engineMode       = store.engineMode
         state.localASRBackend  = store.localASRBackend
-        state.localModelsReady = OnDeviceModelStatus.isLocalStackReady(
-            asrBackend: store.localASRBackend
-        )
-        state.localModelsLoaded = OnDeviceModelStatus.modelsLoadedInMemory()
+        // v0.2.0: iOS `SpeechAnalyzer` is always ready; mirror that
+        // into the State flags so downstream consumers see the same
+        // shape they did when the previous Qwen3 stack reported "ready".
+        state.localModelsReady = true
+        state.localModelsLoaded = false
 
         #if DEBUG
         // Print a masked view of the live App Group config so we can see
@@ -74,10 +75,11 @@ public struct AppGroupPersistor {
         let store = AppGroupStore()
         state.engineMode = store.engineMode
         state.localASRBackend = store.localASRBackend
-        state.localModelsReady = OnDeviceModelStatus.isLocalStackReady(
-            asrBackend: store.localASRBackend
-        )
-        state.localModelsLoaded = OnDeviceModelStatus.modelsLoadedInMemory()
+        // v0.2.0: iOS `SpeechAnalyzer` is always ready. Keep these
+        // toggles here so the keyboard UI doesn't flicker if the host
+        // app briefly clears them while refactoring.
+        state.localModelsReady = true
+        state.localModelsLoaded = false
     }
 
     /// Persist `mode` to the App Group store.