[Feature Request] Frontend to emit i1 packed tensor for attention masks

[lialan]

Request description

As the title says. In #19354 we are adding one extra encoding attribute #iree_encoding.packed_storage to indicate that a tensor has a back to back packed memory layout. such as tensor<1024x1024xi1, #iree_encoding.packed_storage>.

The backend will handle such tensor accordingly.

What component(s) does this issue relate to?

No response

Additional context

No response

[lialan]

@rsuderman #19354 will enable packed tensor types, which the frontend can emit and the backend can process it properly.