[llvm] [AArch64][SVE] Lower scalar FP converts to SVE when Neon is unavailable (PR #112564)

[Sander de Smalen via llvm-commits]

sdesmalen-arm wrote:

> > To me this looks more like a DAGCombine rather than a lowering problem. You're essentially saying `fp_op->gpr->fpr` should be rewritten as `new_fp_op` when possible. I'm assuming there are related combines for the cases where the only reason for the GPR was to either load the source data or store the result, which again feels like a DAG or preferable an IR combine.
> 
> There was some discussion on that here [#112564 (comment)](https://github.com/llvm/llvm-project/pull/112564#discussion_r1803497212), originally this followed the neon combine that matched `(from_int (to_int))` -- [564d9a7](https://github.com/llvm/llvm-project/commit/564d9a78713e1f95dbef6a14441b82d77c25f04d), but it was suggested to make it into a more generally applied lowering.

I hope this wasn't a bum steer, but I thought there might be more general value in this for other cases than this particular fp -> int -> fp combine, e.g. when `int` result is inserted back into a vector for further (vector) calculations, although I didn't have any particular case in mind.

https://github.com/llvm/llvm-project/pull/112564