[llvm] specify NaN behavior more precisely (PR #66579)

[Jacob Lifshay via llvm-commits]

programmerjake wrote:

> I'm pretty sure the significand lives in the least significant bits on every single architecture. For format conversions I'm pretty sure everyone just truncates/zero extends.

IEEE754 specifies fp->fp format conversions should use the programmer-chosen rounding mode, for LLVM for f32 <-> f64 that's always round to nearest, ties to even

https://github.com/llvm/llvm-project/pull/66579