[llvm] [LLVM] Add `llvm.vector.reduce.fdot` intrinsic (PR #192174)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm requested changes to this pull request.

I'm skeptical that an intrinsic is necessary for this. Additionally, fast math flags should not be used as lowering modifiers. 

https://github.com/llvm/llvm-project/pull/192174