[llvm] [LV][EVL] Support icmp/fcmp instruction with EVL-vectorization (PR #108533)

[Philip Reames via llvm-commits]

preames wrote:

JFYI, we've now landed a change which moves away from representing most non-trapping instructions using VP intrinsics, and instead relying on the VLOptimizer to reduce the VL for normal IR nodes instead.  Skimming this review title, I suspect this patch is no longer needed and can be closed.  

https://github.com/llvm/llvm-project/pull/108533