[llvm] [MIPS] match llvm.{min, max}num with {min, max}.fmt for R6 (PR #89021)

[YunQiang Su via llvm-commits]

wzssyqa wrote:

Ohh, this patch is not correct.

In https://llvm.org/docs/LangRef.html#llvm-minnum-intrinsic :
```
Unlike the IEEE-754 2008 behavior, this does not distinguish between signaling and quiet NaN inputs. If a target’s implementation follows the standard and returns a quiet NaN if either input is a signaling NaN, the intrinsic lowering is responsible for quieting the inputs to correctly return the non-NaN input (e.g. by using the equivalent of llvm.canonicalize).
```

https://github.com/llvm/llvm-project/pull/89021