[llvm] [AArch64] Make use of byte FPR stores for bytes extracted from vectors (PR #134117)

[Benjamin Maxwell via llvm-commits]

MacDue wrote:

It might be easier if the generalization of the fold was a separate patch (and this PR lands the previous byte store combine). That way, the test changes due to changing the non-i8 fold would be in their own patch. WDYT?

https://github.com/llvm/llvm-project/pull/134117