[llvm] [AArch64] Extend custom lowering for SVE types in `@llvm.experimental.vector.compress` (PR #105515)

[Lawrence Benson via llvm-commits]

lawben wrote:

I've found a bug during manual testing and I'm not sure yet what the issue is. So I'd put this PR is on hold for now, until I've fixed this. As I'm traveling for ~10 days, this will probably take two weeks or so. 

It looks like when I extract subvectors to use `compact` for larger types and store them back to memory, the vectors are overwriting each other. Not sure if this has something to do with store reordering or something similar.

https://github.com/llvm/llvm-project/pull/105515