[llvm] [AArch64] Fix SVE scalar fcopysign lowering without neon. (PR #129787)

[Paul Walker via llvm-commits]

paulwalker-arm wrote:

Just a FYI but I have recently refactored LowerFCOPYSIGN to simplify it somewhat whilst at the same time making it work better for bfloat types.  Nothing to block this PR, but you can leave the bfloat side of things to me.

https://github.com/llvm/llvm-project/pull/129787