Allow int8 type in quantized_layer_norm (#5899)

Summary:

The meta kernel was returning uint8 as a hardcoded value.

Reviewed By: zonglinpeng

Differential Revision: D63659948

Author: mcremon-meta

backends/cadence/aot/ops_registrations.py

@@ -164,7 +164,7 @@ def quantized_layer_norm_meta(
     output_scale: float,
     output_zero_point: int,
 ) -> torch.Tensor:
-    return input.new_empty(input.size(), dtype=torch.uint8)
+    return input.new_empty(input.size(), dtype=input.dtype)
 
 
 @register_fake("cadence::quantized_relu")