vcvt_u32_f32 saturating behaviour

[cwiede]

Hi, first of all thanks for this great project. It's working like a charm!

I use the library in this way for simulating arm intrinsics on Intel/SSE machines.

#define SIMDE_ENABLE_NATIVE_ALIASES 
#include "simde/arm/neon.h"

Almost everything works perfectly, there is just a minor issue with saturating semantics of vcvt_u32_f32 and similar functions. These are defined to use the round towards zero mode, and so they are expected to saturate (and indeed they do on the ARM HW).

It is unfortunately not true with simde:

vcvt_u32_f32([2.1474836e+09, 4.2949673e+09])

is expected to return

[2147483648 4294967295]

but it actually returns

[2147483648 0]

Similar things happen with other vcvt variants and also the vcvt_n variants.

[mr-c]

@cwiede You are very welcome. Can you send a pull request to fix this?