feat: initial public release v0.1.0 — MLX port of pyannote-speaker-diarization-3.1

Byte-parity with pyannote-PyTorch reference (cosine 0.763718 identical
at 6 decimals on 200 cross-window slot pairs). 2.5x faster than
pyannote-MPS on Apple Silicon native.

Extracted from gitea.tavportal.com/olivier/MLX_CONVERTOR commit 5f9eafa.

tests/unit/test_diar_window.py

@@ -0,0 +1,16 @@
+from pyannote_diarization_3_1_mlx._window import sliding_windows
+import numpy as np
+
+
+def test_sliding_windows_full_coverage():
+    sr = 16000
+    audio = np.zeros(int(25 * sr), dtype=np.float32)
+    windows = list(sliding_windows(audio, sr=sr, duration_s=10.0, hop_s=1.0))
+    # Expect (25-10)/1 + 1 = 16 windows, all 10 s long
+    assert len(windows) == 16
+    for start, end, slice_ in windows:
+        assert end - start == 10.0
+        assert len(slice_) == 10 * sr
+    # boundaries
+    assert windows[0][0] == 0.0
+    assert windows[-1][1] == 25.0