feat(workers): whisper loader with CUDA detect + CPU fallback

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

workers/tests/test_model.py

@@ -0,0 +1,23 @@
+from unittest.mock import patch, MagicMock
+from void_workers import model
+
+
+def test_model_returns_singleton(monkeypatch):
+    m = MagicMock()
+    monkeypatch.setattr(model, "_whisper_model", None)
+    with patch("void_workers.model.cuda_available", return_value=False):
+        with patch("faster_whisper.WhisperModel", return_value=m):
+            a = model.whisper_model()
+            b = model.whisper_model()
+            assert a is b
+
+
+def test_transcribe_returns_joined_segments(monkeypatch):
+    seg1 = MagicMock(text=" Hello world ")
+    seg2 = MagicMock(text=" second line")
+    fake_model = MagicMock()
+    fake_model.transcribe.return_value = ([seg1, seg2], MagicMock())
+    monkeypatch.setattr(model, "_whisper_model", fake_model)
+    out = model.whisper_transcribe("/tmp/whatever.opus")
+    assert "Hello world" in out
+    assert "second line" in out