tf.aliasing support

[steeve]

Hi,

We (@zml) found that tf.aliasing support seemed to be not working as expected, with the model producing garbage when used. In our case Llama 3.1 8B.
This is problematic for transformer models because we leverage donations for the KvCache.

For now we're not emitting those attributes when on neuron, but we're not sure what to do as we feel that if the SDK doesn't support them, it should just ignore them right ?

The llama implementation is attached.

Thank you !

llama.aliasing.mlir.txt