Enable NoNaNs fast math flag

[adazem009]

Enabling this flag results in a nice performance improvement. However, it breaks NaN detection. The flag can still be enabled if we ignore it in LLVMBuildUtils::isNaN(). This only possible since LLVM 20.1 (https://github.com/llvm/llvm-project/commit/a77346bad0d80a7f7e184f3d6e12c7f532101136) which introduces CreateFCmpFMF().