[llvm] [ScalarizeMaskedMemIntr] Don't use a scalar mask on GPUs (PR #104842)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

For context, this pass runs on the LLVM IR side of codegen and is responsible for taking `llvm.masked.*` to scalar code if those masked loads aren't legal on a given target so we don't have everyone writing duplicate logic down in MIR legalization (is what I've gathered from the git history)

https://github.com/llvm/llvm-project/pull/104842