Wakeword calibration and deployment.

Summary: Create a separate stateful wakeword model that keeps states as buffers. This avoids the need to explicitly manage/copy states between execution.

Reviewed By: mcremon-meta

Differential Revision: D68301862

Author: hsharma35

backends/cadence/aot/compiler.py

@@ -211,6 +211,7 @@ def export_to_edge(
                 torch.ops.aten.linalg_vector_norm.default,
                 torch.ops.aten.unfold.default,
                 torch.ops.aten.angle.default,
+                torch.ops.aten.empty_permuted.default,
             ],
         ),
     )