[clang] [HIP] Remove redundant device optimization pipeline (PR #206350)

[Joseph Huber via cfe-commits]

https://github.com/jhuber6 commented:

What's the motivation for this? Trying to cut down on compilation times by removing one of the passes? I'd expect that to cause massive performance regressions because the LTO pipeline expects to run against optimized IR, and I'd prefer we don't add more things that keep us doing the current `-x ir` hack, I'd like that to be gone before release at a minimum.

https://github.com/llvm/llvm-project/pull/206350