Fix Web Speech API desktop voice recognition

- Set continuous=true to keep listening through pauses
- Only process final results, ignore interim transcripts
- Add usesFallback check to route Web Speech API transcripts through classification
- Desktop now captures complete phrases before classification
- Add detailed logging for debugging recognition flow

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

maternal-web/components/voice/VoiceFloatingButton.tsx

@@ -44,10 +44,10 @@ export function VoiceFloatingButton() {
     severity: 'info',
   });
 
-  const { isListening, isSupported, transcript, classification, error, startListening, stopListening, reset } =
+  const { isListening, isSupported, transcript, classification, error, usesFallback, startListening, stopListening, reset } =
     useVoiceInput();
 
-  // Auto-use classification from backend when transcription completes
+  // Auto-use classification from backend when transcription completes (MediaRecorder fallback)
   React.useEffect(() => {
     if (classification && !isListening && !isProcessing && open) {
       setClassificationResult(classification);
@@ -55,6 +55,13 @@ export function VoiceFloatingButton() {
     }
   }, [classification, isListening, isProcessing, open]);
 
+  // For Web Speech API (desktop), classify the transcript client-side
+  React.useEffect(() => {
+    if (!usesFallback && transcript && !isListening && !isProcessing && open && transcript !== lastClassifiedTranscript) {
+      classifyTranscript(transcript);
+    }
+  }, [usesFallback, transcript, isListening, isProcessing, open, lastClassifiedTranscript]);
+
   const handleOpen = () => {
     if (!isSupported) {
       setSnackbar({