[llvm] [RISCV] Use vmv.v.x to materialize masks in deinterleave2 lowering (PR #118500)

[Philip Reames via llvm-commits]

preames wrote:

A bit of context for reviewers - I was originally thinking that we'd either use an i8 vid sequence (which since most of the cases flowing through this would be e64 would only be a vtype toggle), or choice sequences based on LMUL, but I realized that a) small LMUL here is pretty uncommon because an m1 result still requires an m2 vcompress, b) the complexity just didn't seem worth it, and c) the vmv.v.x can be materialized while the vid/vadd/vmsne can't.  

https://github.com/llvm/llvm-project/pull/118500