[llvm] [LLVM] Add `llvm.masked.compress` intrinsic (PR #92289)

[Lawrence Benson via llvm-commits]

lawben wrote:

I've added a [placeholder PR](https://github.com/lawben/llvm-project/pull/1) in my fork for SVE lowering on AArch64, so you can see how I will continue.  Next steps would be custom lowering for x86 and RISCV. And I would like to do a target-independent compress+store combine, to avoid the temporary stack stores of we just load+store again.

Question to all: What is missing for this PR to be merged? I guess the GlobalISel stuff is still open (@arsenm), but I need some input there. Other than that, what do I still need to implement/do?    

https://github.com/llvm/llvm-project/pull/92289