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

[Joseph Huber via llvm-commits]

jhuber6 wrote:

> FWIW, the pass should be super cheap, if it starts by looking for the intrinsic and then the uses. Running it twice is a valid option.

First pass should delete all the uses, the concern is that by not trimming the intrinsic earlier we're losing some optimizations, but I feel like we're still getting a full optimization pipeline and this likely will be easily optimized out. If @Artem-B is really concerned I'll just change it to keep the per-file run but ignore it if there's no SM passed.

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