[llvm] [LoopVectorize] Support vectorization of compressing patterns (PR #214491)

[Benjamin Maxwell via llvm-commits]

MacDue wrote:

> I think `-mtriple riscv64 -mattr=+v` should be enough for RISC-V

That's what I used, though as noted above we also need `-force-vector-width=4` as RISC-V seems to prefer scalable VFs and scalable compressstores/expandloads are reported as illegal.  

https://github.com/llvm/llvm-project/pull/214491