[llvm] [NVPTX] Do not run the NVVMReflect pass as part of the normal pipeline (PR #121834)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

I think I could probably make @Artem-B happy if I just forewent the early pass if the architecture is not known. That'd leave CUDA with identical behavior while allowing this kind of use where we only specify the target during the final link + backend stage.

https://github.com/llvm/llvm-project/pull/121834