util: Keep quiet NaNs quiet when converting to half float.
We don't want to be throwing exceptions and changing float values later by emitting a signaling binary16 nan. If we don't do this, then when we convert back to f32 in NIR constant expression evaluation, the signaling NaN can end up giving NaN for fmax(NaN, 0.0), instead of 0.0. Closes: mesa/mesa#5933 Cc: mesa-stable Reviewed-by:Emma Anholt <emma@anholt.net> Reviewed-by:
Jason Ekstrand <jason.ekstrand@collabora.com> Part-of: <mesa/mesa!16233>