Use local faster-whisper only for transcription.

Drop cloud and Ollama engine choices so audio stays on-machine via Docker Whisper, and tighten orphan detection plus UI around a single local flow.

app/Http/Controllers/TranscribeController.php

@@ -10,9 +10,9 @@ use Illuminate\Http\RedirectResponse;
 class TranscribeController extends Controller
 {
     /**
-     * Queue transcription for the recording with the chosen engine.
+     * Queue local faster-whisper transcription for the recording.
      *
-     * Always allowed: stops any current run first, then starts the new engine.
+     * Always allowed: stops any current run first, then starts a new one.
      */
     public function __invoke(TranscribeRecordingRequest $request, Recording $recording): RedirectResponse
     {
@@ -21,11 +21,9 @@ class TranscribeController extends Controller
             $recording->refresh();
         }
 
-        $driver = $request->validated('driver');
-
         $recording->update([
-            'transcription_driver' => $driver,
-            'ollama_url' => $driver === 'ollama' ? rtrim($request->validated('ollama_url'), '/') : null,
+            'transcription_driver' => 'local',
+            'ollama_url' => null,
             'transcription_status' => 'pending',
             'transcription_progress' => 'Queued — waiting to start…',
             'transcription_percent' => 5,