[llvm] [AArch64] Enable FEAT_LSFE CodeGen only when exception can be ignored (PR #212799)

[via llvm-commits]

Lukacma wrote:

I have marked it as draft, since I am not sure whether this is the right approach to take. I have [RFC](https://discourse.llvm.org/t/rfc-add-fast-math-flags-to-atomicrmw/91411) discussion on similar topic, but there the argument for guarding the codegen is based around NaN propagation which is not as important to users as exception handling. Therefore this feels less controversial to 

https://github.com/llvm/llvm-project/pull/212799