[Feature]: confirm nothing depends on the 0-dim device index constraint

[yannicks1]

Feature description

test_spyre_indirect_matmul_tensor_index was xfail(strict=True) and started passing with the
torch-spyre bump to 9f975a3 (#950): a 0-dim scalar device index feeding transpose into matmul
now matches CPU.

Confirm whether any code was shaped by that constraint and can be removed or simplified.

Motivation and context

If a simplification is found, benchmark prefill and decode before and after.

Proposed solution

No response

Checklist

• I have searched for similar feature requests