[llvm] [AArch64] Fix #94909: Optimize vector fmul(sitofp(x), 0.5) -> scvtf(x, 2) (PR #141480)

[Stephen Canon via llvm-commits]

stephentyrone wrote:

Looks good, thanks for your persistence. Can you add a couple of test cases for examples where we _can't_ do the transform because the power of two is out of range?

https://github.com/llvm/llvm-project/pull/141480