fix(dictation): handle transcription failed events

OpenAI emits conversation.item.input_audio_transcription.failed when an
item cannot be transcribed (e.g. audio_unintelligible). Previously this
event was dropped by the type filter, leaving isTranscribing stuck true
with no user feedback after a preceding buffer commit.

Now the failed event is treated as terminal: clears the transcribing
indicator and surfaces the error message via toast.

desktop/src/features/dictation/hooks/useRealtimeDictation.ts

@@ -8,6 +8,7 @@ import {
   BUFFER_COMMITTED_EVENT,
   TRANSCRIPT_COMPLETED_EVENT,
   TRANSCRIPT_DELTA_EVENT,
+  TRANSCRIPT_FAILED_EVENT,
   commitAudioBuffer,
   connectPeerConnection,
   createAudioBufferCapture,
@@ -185,6 +186,13 @@ export function useRealtimeDictation({
         return;
       }
 
+      if (event.type === TRANSCRIPT_FAILED_EVENT) {
+        console.warn("OpenAI transcription failed for item", event);
+        setIsTranscribing(false);
+        toast.error(event.error?.message ?? "Transcription failed");
+        return;
+      }
+
       if (
         event.type !== TRANSCRIPT_DELTA_EVENT &&
         event.type !== TRANSCRIPT_COMPLETED_EVENT &&

desktop/src/features/dictation/lib/realtimeAudio.ts

@@ -8,6 +8,8 @@ export const TRANSCRIPT_DELTA_EVENT =
   "conversation.item.input_audio_transcription.delta";
 export const TRANSCRIPT_COMPLETED_EVENT =
   "conversation.item.input_audio_transcription.completed";
+export const TRANSCRIPT_FAILED_EVENT =
+  "conversation.item.input_audio_transcription.failed";
 export const BUFFER_COMMITTED_EVENT = "input_audio_buffer.committed";
 
 const MAX_BUFFER_CHUNKS = 500; // ~10s at 20ms per chunk