Arm backend: Add mean.dim to CheckProperQuantization (pytorch#12127)

mean.dim decomposes into an avg_pool2d, which occurs after quantization
annotation. This led to the avg_pool2d node not having quantization
parameters for the output. mean.dim is added to CheckProperQuantization
so that if the decomposition occurs the avg_pool2d operator will not be
delegated.

Signed-off-by: Sebastian Larsson <[email protected]>

Author: Sebastian-Larsson

backends/arm/operator_support/tosa_supported_operators.py

@@ -327,6 +327,7 @@ class CheckProperQuantization(OperatorSupportBase):
         exir_ops.edge.aten.upsample_bilinear2d.vec,
         exir_ops.edge.aten.upsample_nearest2d.vec,
         torch.ops.aten.scalar_tensor.default,
+        exir_ops.edge.aten.mean.dim,
         *TableOps.included_ops(),
     )