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/FlowAppLifecycle.swift

@@ -0,0 +1,39 @@
+// FlowAppLifecycle.swift
+// OSGKeyboard · Shared
+//
+// Tracks whether the host app process is in the foreground.
+// Retained for any future GPU-backed paths; CoreML ASR does not require it.
+
+import Foundation
+
+public final class FlowAppLifecycle: @unchecked Sendable {
+
+    public static let shared = FlowAppLifecycle()
+
+    private let lock = NSLock()
+    private var isForeground = true
+
+    private init() {}
+
+    /// `true` when the host app scene is active (`.active`).
+    public var allowsGPUInference: Bool {
+        lock.lock()
+        defer { lock.unlock() }
+        return isForeground
+    }
+
+    public func setForeground(_ foreground: Bool) {
+        lock.lock()
+        isForeground = foreground
+        lock.unlock()
+    }
+
+    /// Blocks until foreground or cancellation.
+    public func waitUntilForeground() async -> Bool {
+        while !allowsGPUInference {
+            if Task.isCancelled { return false }
+            try? await Task.sleep(nanoseconds: 200_000_000)
+        }
+        return true
+    }
+}