fix(voice,knowledge): mic recording + pagination + STT-only mode

Voice: recreate AudioRecorder each session to avoid stale native
state, improve permission handling (re-check after settings), show
feedback after 3 consecutive empty transcripts, allow STT-only mode
when TTS is unavailable.

Knowledge: hydrate IDs immediately after loading more pages so
scroll-based pagination doesn't stall at the bottom.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

lib/data/api/voice_api.dart

@@ -15,8 +15,9 @@ class VoiceStatus {
     required this.tts,
   });
 
-  /// True only when voice is enabled AND both STT and TTS are ready.
-  bool get fullyAvailable => enabled && stt && tts;
+  /// True when voice is enabled and at least STT is ready.
+  /// TTS is optional — voice mode works without it (STT-only).
+  bool get fullyAvailable => enabled && stt;
 
   factory VoiceStatus.fromJson(Map<String, dynamic> json) => VoiceStatus(
         enabled: json['enabled'] as bool? ?? false,

lib/providers/knowledge_provider.dart

@@ -246,6 +246,9 @@ class KnowledgeNotifier extends Notifier<KnowledgeState> {
         isLoadingIds: false,
         hasMore: combined.length < total,
       );
+      // Hydrate the newly fetched IDs immediately — the user is
+      // already at the scroll bottom so _onScroll won't re-fire.
+      await hydrateNext();
     } catch (_) {
       state = state.copyWith(isLoadingIds: false);
     }

lib/providers/voice_provider.dart

@@ -112,6 +112,9 @@ class VoiceNotifier extends Notifier<VoiceState> {
   // to reduce STT mishearings of domain-specific words.
   String _lastAssistantContent = '';
 
+  // Empty transcript counter — show feedback after consecutive blanks
+  int _emptyTranscriptCount = 0;
+
   // TTS playback queue
   final _ttsQueue = Queue<Uint8List>();
   bool _ttsPlaying = false;
@@ -120,7 +123,6 @@ class VoiceNotifier extends Notifier<VoiceState> {
 
   @override
   VoiceState build() {
-    _recorder = AudioRecorder();
     _player = AudioPlayer();
     ref.onDispose(() {
       _amplitudeSubscription?.cancel();
@@ -146,32 +148,38 @@ class VoiceNotifier extends Notifier<VoiceState> {
       return;
     }
 
-    // Check server availability
+    // Check server availability — STT is required, TTS is optional.
     try {
       final status = await ref.read(voiceRepositoryProvider).checkStatus();
-      if (!status.fullyAvailable) {
-        onError('Voice not available on this server');
+      if (!status.enabled || !status.stt) {
+        onError('Speech-to-text not available on this server');
         return;
       }
+      // Downgrade to STT-only when TTS is unavailable
+      if (!status.tts) enableTts = false;
     } catch (_) {
-      onError('Voice not available on this server');
+      onError('Could not reach voice service');
       return;
     }
 
     // Check microphone permission
-    final permStatus = await Permission.microphone.request();
-    if (permStatus == PermissionStatus.denied ||
-        permStatus == PermissionStatus.permanentlyDenied) {
+    var permStatus = await Permission.microphone.request();
+    if (permStatus == PermissionStatus.permanentlyDenied) {
+      onError('Microphone blocked — opening settings');
+      final opened = await openAppSettings();
+      if (!opened) return;
+      // Re-check after user returns from settings
+      permStatus = await Permission.microphone.status;
+    }
+    if (!permStatus.isGranted) {
       onError('Microphone permission required');
-      if (permStatus == PermissionStatus.permanentlyDenied) {
-        await openAppSettings();
-      }
       return;
     }
 
     _onTranscript = onTranscript;
     _onError = onError;
     _enableTts = enableTts;
+    _emptyTranscriptCount = 0;
     _tempDir = await getTemporaryDirectory();
 
     state = state.copyWith(voiceModeActive: true, available: true);
@@ -225,12 +233,22 @@ class VoiceNotifier extends Notifier<VoiceState> {
     state = state.copyWith(mode: VoiceMode.recording);
 
     final dir = _tempDir ?? await getTemporaryDirectory();
-    // AAC/M4A is reliably supported on Android (unlike WebM/Opus which can
-    // produce OGG bytes in a .webm file, confusing server-side decoders).
     final path =
         '${dir.path}/voice_rec_${DateTime.now().millisecondsSinceEpoch}.m4a';
 
     try {
+      // Recreate the recorder each session — the record package can leave
+      // the native AudioRecord in a bad state after stop/error cycles,
+      // and a stale instance is the most common cause of "could not start".
+      _recorder?.dispose();
+      _recorder = AudioRecorder();
+
+      if (!await _recorder!.hasPermission()) {
+        _onError?.call('Microphone permission was revoked');
+        exitVoiceMode();
+        return;
+      }
+
       await _recorder!.start(
         const RecordConfig(encoder: AudioEncoder.aacLc, sampleRate: 16000),
         path: path,
@@ -241,7 +259,7 @@ class VoiceNotifier extends Notifier<VoiceState> {
           .onAmplitudeChanged(const Duration(milliseconds: 200))
           .listen(_onAmplitude);
     } catch (e) {
-      _onError?.call('Microphone error: could not start recording');
+      _onError?.call('Microphone error: $e');
       exitVoiceMode();
     }
   }
@@ -291,10 +309,16 @@ class VoiceNotifier extends Notifier<VoiceState> {
       if (!state.voiceModeActive) return;
 
       if (transcript.isEmpty) {
-        // Empty transcript — restart silently
+        _emptyTranscriptCount++;
+        if (_emptyTranscriptCount >= 3) {
+          _onError?.call('No speech detected — tap the mic to try again');
+          exitVoiceMode();
+          return;
+        }
         await _startListening();
         return;
       }
+      _emptyTranscriptCount = 0;
 
       // Reset TTS state for this new turn
       _sentenceBuffer = '';