[llvm] [AArch64] Add scal_to_vec patterns for rounding nodes (PR #173004)

[via llvm-commits]

Lukacma wrote:

The only way I can imagine that working is if we convert scal_to_vec into  v1i64  to bitcast during lowering. I didn't go that route as I wasn't sure what kind of knock off effects such change would have on rest of the backend, but I will give it a shot and see.

https://github.com/llvm/llvm-project/pull/173004