[llvm] [Mips] Fix wrong qNaN encoding when -mnan=legacy (PR #153777)

[Nikita Popov via llvm-commits]

nikic wrote:

It might make sense to start a discussion on discourse on what to do with legacy MIPS NaN. What you do here is going to be a very partial solution, because we have tons of code that assumes IEEE NaNs.

I'm not even sure what it would take to really support this. I guess the proper way would be to add a new APFloat fltSemantics for legacy MIPS floats, but then it's unclear how those would map to the IR floating point types.

We should probably also consider making `-mnan=legacy` an error and stop pretending that we support it.

https://github.com/llvm/llvm-project/pull/153777