feat(stt): pass last assistant response as Whisper context to reduce mishearings

lib/providers/voice_provider.dart

@@ -108,6 +108,10 @@ class VoiceNotifier extends Notifier<VoiceState> {
   int _lastSeenLength = 0;
   bool _streamComplete = false;
 
+  // Last complete assistant response — passed to Whisper as initial_prompt
+  // to reduce STT mishearings of domain-specific words.
+  String _lastAssistantContent = '';
+
   // TTS playback queue
   final _ttsQueue = Queue<Uint8List>();
   bool _ttsPlaying = false;
@@ -205,6 +209,7 @@ class VoiceNotifier extends Notifier<VoiceState> {
     _dispatchSentences(flush: isComplete);
 
     if (isComplete) {
+      _lastAssistantContent = fullContent;
       _streamComplete = true;
       _checkRestartListening();
     }
@@ -278,8 +283,10 @@ class VoiceNotifier extends Notifier<VoiceState> {
 
       if (!state.voiceModeActive) return;
 
-      final transcript =
-          await ref.read(voiceRepositoryProvider).transcribe(bytes);
+      final transcript = await ref.read(voiceRepositoryProvider).transcribe(
+            bytes,
+            context: _lastAssistantContent.isNotEmpty ? _lastAssistantContent : null,
+          );
 
       if (!state.voiceModeActive) return;