Defer AVSpeechSynthesisVoice init to first speak() call

AVSpeechSynthesisVoice(language:) triggers a malloc double-free on
iOS 26 simulators when deserializing voice metadata during app launch.
Move voice resolution from init() to first speak() so the framework
call happens after the app is fully initialized.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Conjuga/Conjuga/Services/SpeechService.swift

@@ -4,14 +4,20 @@ import AVFoundation
 @MainActor
 final class SpeechService {
     private let synthesizer = AVSpeechSynthesizer()
-    private let spanishVoice: AVSpeechSynthesisVoice?
+    private var spanishVoice: AVSpeechSynthesisVoice?
+    private var voiceResolved = false
     private var audioSessionConfigured = false
 
     init() {
-        spanishVoice = AVSpeechSynthesisVoice(language: "es-ES")
+        // AVSpeechSynthesisVoice can trigger a malloc double-free on
+        // iOS 26 simulators when deserializing voice metadata. Defer
+        // voice resolution to first use so the crash doesn't happen
+        // during app launch.
+        spanishVoice = nil
     }
 
     func speak(_ text: String) {
+        resolveVoiceIfNeeded()
         configureAudioSession()
         synthesizer.stopSpeaking(at: .immediate)
         let utterance = AVSpeechUtterance(string: text)
@@ -27,6 +33,12 @@ final class SpeechService {
         synthesizer.stopSpeaking(at: .immediate)
     }
 
+    private func resolveVoiceIfNeeded() {
+        guard !voiceResolved else { return }
+        voiceResolved = true
+        spanishVoice = AVSpeechSynthesisVoice(language: "es-ES")
+    }
+
     private func configureAudioSession() {
         guard !audioSessionConfigured else { return }
         do {