[llvm] Changes to make Float-to-int scalar transform codegen deterministic (PR #92551)

[Arthur Eubanks via llvm-commits]

aeubanks wrote:

I've successfully used llvm-reduce to reduce non-determinism from a large module with an interestingness test that optimizes ~20 times and checks if outputs differ

https://github.com/llvm/llvm-project/pull/92551