[llvm] [LLVM][NVPTX]Add BF16 vector instruction and fix lowering rules (PR #69415)

[Thomas Raoux via llvm-commits]

https://github.com/ThomasRaoux approved this pull request.

LGTM, just added a comment about why select depends on sm_90 which looks surprising to me.

https://github.com/llvm/llvm-project/pull/69415